Вы находитесь на странице: 1из 137

Guidelines on the Evaluation of

Vector Network Analysers (VNA)


EURAMET Calibration Guide No. 12
Version 3.0 (03/2018)
Authorship and Imprint
This document was developed by the EURAMET e.V., Technical Committee for Electricity and Magnetism.
Authors: Markus Zeier (METAS, Switzerland), Djamel Allal (LNE, France), Rolf Judaschke (PTB, Germany).

Acknowledgement
The authors would like to thank for reviewing the guide: Thomas Reichel (Technical consultant), Blair Hall
(MSL), Gary Bennett (National Instruments), Dave Blackham (Keysight Technologies), Andreas C. Böck
(esz AG calibration & metrology), Andy Brush (TEGAM), Tekamul Buber (Maury Microwave), Albert Calvo
(Rohde & Schwarz), Onur Cetiner (Keysight Technologies), Chris Eio (NPL), Andrea Ferrero (Keysight
Technologies), Israel Garcia Ruiz (CENAM), Martin Grassl (Spinner), Tuomas Haitto (Millog Oy), Johannes
Hoffmann (METAS), Matthias Hübler (Rohde & Schwarz), Ian Instone (Technical consultant), Harald Jäger
(Rohde & Schwarz), Karsten Kuhlmann (PTB), Jian Liu (Keysight Technologies), Linoh Magalula (NMISA),
Jon Martens (Anritsu), Guillermo Monasterios (INTI), Faisal Mubarak (VSL), Rusty Myers (Keysight
Technologies), Reiner Oppelt (Rosenberger), Nick Ridler (NPL), Juerg Ruefenacht (METAS), Handan
Sakarya (UME), Bart Schrijver (Keysight Technologies), Joachim Schubert (Rosenberger), Nosherwan
Shoaib (INRIM, NUST), Hernando Silva (INTI), Pamela Silwana (NMISA), Laszlo Sleisz (NMHH), Daniel
Stalder (METAS), Michael Wollensack (METAS), Ken Wong (Keysight Technologies), Sherko Zinal (PTB).

Version 3.0 March 2018


Version 2.0 March 2011
Version 1.0 July 2007

EURAMET e.V.
Bundesallee 100
38116 Braunschweig
Germany

E-mail: secretariat@euramet.org
Phone: +49 531 592 1960

Official language
The English language version of this document is the definitive version. The EURAMET Secretariat can
give permission to translate this text into other languages, subject to certain conditions available on
application. In case of any inconsistency between the terms of the translation and the terms of this
document, this document shall prevail.

Copyright
The copyright of this document (EURAMET Calibration Guide No. 12, version 3.0 – English version) is held
by © EURAMET e.V. 2007. The text may not be copied for sale and may not be reproduced other than in
full. Extracts may be taken only with the permission of the EURAMET Secretariat.

ISBN 978-3-942992-51-0

Image on cover page by PTB.

Guidance publications
This document gives guidance on measurement practices in the specified fields of measurements. By
applying the recommendations presented in this document, laboratories can produce calibration results
that can be recognised and accepted throughout Europe. The approaches taken are not mandatory and
are for the guidance of calibration laboratories. The document has been produced as a means of promoting
a consistent approach to good measurement practice leading to and supporting laboratory accreditation.

The guide may be used by third parties, e.g. National Accreditation Bodies, peer reviewers, witnesses to
measurements, etc., as a reference only. Should the guide be adopted as part of a requirement of any such
party, this shall be for that application only and the EURAMET Secretariat should be informed of any such
adoption.

EURAMET Calibration Guide No. 12 I-CAL-GUI-012/v3.0/2018-03-14


Version 3.0 (03/2018)
On request EURAMET may involve third parties in stakeholder consultations when a review of the guide is
planned. If you are interested, please contact the EURAMET Secretariat.

No representation is made nor warranty given that this document or the information contained in it will be
suitable for any particular purpose. In no event shall EURAMET, the authors or anyone else involved in the
creation of the document be liable for any damages whatsoever arising out of the use of the information
contained herein. The parties using the guide shall indemnify EURAMET accordingly.

Further information
For further information about this document, please contact your national contact person in the EURAMET
Technical Committee for Electricity and Magnetism (see https://www.euramet.org)

EURAMET Calibration Guide No. 12 I-CAL-GUI-012/v3.0/2018-03-14


Version 3.0 (03/2018)
EURAMET Calibration Guide No. 12

Version 3.0 (03/2018)

Guidelines on the Evaluation of Vector


Network Analysers (VNA)

Purpose
This document has been produced to enhance the equivalence and mutual recognition of
calibration results obtained by laboratories performing measurements with vector network
analysers.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018)
Contents

List of Figures 6

List of Tables 7

1 Introduction 9
1.1 Purpose of this guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Comparison with previous guideline . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Scope and Applicability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.1 VNA calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.2 Error model and measurement model . . . . . . . . . . . . . . . . . . . 11
1.4.3 Error coefficients and residual errors . . . . . . . . . . . . . . . . . . . . 11

2 Traceability schemes and measurement standards 11


2.1 Traceability chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Measurement standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 VNA calibration 13
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Practical advice for VNA calibration . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Electronic calibration units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4 Verification 15
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Purpose of verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3 Verification Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3.1 Coincidence tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3.2 Plausibility tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.4 Verification procedure and practical advice . . . . . . . . . . . . . . . . . . . . . 16
4.5 Verification criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

5 Uncertainty contributions 18
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Identification of influence quantities . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Characterization of uncertainty contributions . . . . . . . . . . . . . . . . . . . . 19
5.3.1 Characterization of calibration standards . . . . . . . . . . . . . . . . . . 19
5.3.2 Noise floor and trace noise . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.3.3 VNA non-linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.3.4 VNA drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3.5 Isolation (cross-talk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3.6 Test port cable stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3.7 Connection repeatability . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

6 Measurement model 23

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –2–
7 Uncertainty evaluation 23
7.1 Rigorous uncertainty evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.2 Ripple Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

8 Best measurement practice and practical advice 26


8.1 Environmental conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
8.2 VNA architecture, performance and settings . . . . . . . . . . . . . . . . . . . . 27
8.3 Use of mechanical calibration and verification standards . . . . . . . . . . . . . 28
8.4 Use of electronic calibration units . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.5 Test port cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.5.1 Fixture and layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.5.2 Test port adapters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
8.6 Connector care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.7 Initial stability test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.8 Repeatability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

References 33

Annexes 39

A Glossary 39

B Notation 43

C S-parameter measurement uncertainties 44


C.1 S-parameter data format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
C.2 Representation of measurement uncertainties . . . . . . . . . . . . . . . . . . . 44
C.2.1 Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
C.3 Evaluation of measurement uncertainty . . . . . . . . . . . . . . . . . . . . . . 45
C.4 Uncertainty propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
C.5 Sample statistics and Type A uncertainty . . . . . . . . . . . . . . . . . . . . . . 47
C.5.1 Sample statistics of scalar quantities . . . . . . . . . . . . . . . . . . . . 47
C.5.2 Sample statistics of complex-valued quantities . . . . . . . . . . . . . . 47
C.5.3 Type A uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
C.6 Expanded uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

D VNA calibration 51
D.1 One-port calibration techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
D.2 Two-port calibration techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
D.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
D.2.2 Ten-term error model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
D.2.3 Seven-term error model . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
D.3 Over-determined calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –3–
E VNA verification 57
E.1 Verification standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
E.1.1 One-port verification standards . . . . . . . . . . . . . . . . . . . . . . . 57
E.1.2 Two-port verification standards . . . . . . . . . . . . . . . . . . . . . . . 58
E.2 Quantitative verification criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
E.2.1 Scalar case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
E.2.2 Multivariate case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
E.2.3 Examples of quantitative verification . . . . . . . . . . . . . . . . . . . . 64

F VNA measurement models 67


F.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
F.2 One-port measurement model . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
F.3 Two-port measurement models . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
F.4 N-port measurement model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
F.5 Measurement model based on residual error coefficients . . . . . . . . . . . . . 69
F.6 Uncertainty contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

G Characterization procedures of uncertainty contributions 73


G.1 Noise floor and trace noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
G.2 VNA non-linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
G.3 VNA drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
G.4 Test port cable stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
G.5 Connector repeatability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

H Ripple Method 91
H.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
H.2 Uncertainties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
H.3 Practical Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
H.4 Measurement model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
H.4.1 One-port equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
H.4.2 Two-port equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
H.5 Uncertainty contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
H.5.1 Directivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
H.5.2 Source match . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
H.5.3 Reflection tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
H.5.4 Transmission tracking and load match . . . . . . . . . . . . . . . . . . . 100
H.5.5 Isolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
H.5.6 Drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
H.5.7 Test port cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
H.5.8 Non-linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
H.5.9 Repeatability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

I Waveguide measurements 102


I.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
I.2 Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
I.2.1 VNA test ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –4–
I.2.2 Calibration kits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
I.2.3 Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
I.3 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
I.3.1 SSL calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
I.3.2 TRL calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
I.4 Uncertainty Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
I.4.1 Ripple Assessments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
I.4.2 Other uncertainty components . . . . . . . . . . . . . . . . . . . . . . . 105

J Examples 107
J.1 One-port matched load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
J.2 One-port mismatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
J.3 One-port short . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
J.4 Two-port Adapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
J.5 Two-port 20 dB attenuation device . . . . . . . . . . . . . . . . . . . . . . . . . 122
J.6 Two-port 50 dB attenuation device . . . . . . . . . . . . . . . . . . . . . . . . . 128

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –5–
List of Figures
8.1 VNA test port cable setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
D.1 One-port error model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
D.2 Four-receiver vs. three receiver VNA architecture . . . . . . . . . . . . . . . 53
D.3 Ten-term forward VNA error model . . . . . . . . . . . . . . . . . . . . . . . 54
D.4 Ten-term reverse VNA error model . . . . . . . . . . . . . . . . . . . . . . . 54
D.5 Seven-term VNA error model . . . . . . . . . . . . . . . . . . . . . . . . . . 56
E.1 Schematic Beatty Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
E.2 Simulated Beatty line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
E.3 T-checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
E.4 Quantitative verification criteria for a T-Checker . . . . . . . . . . . . . . . . 65
E.5 Quantitative verification criteria for a matched load . . . . . . . . . . . . . . 66
F.1 One-port VNA measurement model . . . . . . . . . . . . . . . . . . . . . . . 68
F.2 Ten-term forward VNA measurement model . . . . . . . . . . . . . . . . . . 70
F.3 Seven-term VNA measurement model . . . . . . . . . . . . . . . . . . . . . 70
F.4 Residual one-port VNA measurement model . . . . . . . . . . . . . . . . . . 71
F.5 Cable and Connector Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
G.1 Characterization of VNA noise floor . . . . . . . . . . . . . . . . . . . . . . . 75
G.2 Characterization of VNA noise floor at low frequencies . . . . . . . . . . . . 75
G.3 Characterization of VNA trace noise magnitude . . . . . . . . . . . . . . . . 76
G.4 Characterization of VNA trace noise magnitude at low frequencies . . . . . 76
G.5 Characterization of VNA trace noise phase . . . . . . . . . . . . . . . . . . . 77
G.6 Characterization of VNA trace noise phase at low frequencies . . . . . . . . 77
G.7 Characterization of VNA non-linearity: full attenuation range . . . . . . . . . 79
G.8 Characterization of VNA non-linearity: limited attenuation range . . . . . . . 79
G.9 Characterization of VNA non-linearity: frequency dependence . . . . . . . . 80
G.10 Characterization of VNA drift related to reflection . . . . . . . . . . . . . . . 83
G.11 Characterization of VNA drift related to transmission magnitude . . . . . . . 84
G.12 Characterization of VNA drift related to transmission phase . . . . . . . . . 84
G.13 Characterization of VNA test port cable: transmission magnitude stability . . 87
G.14 Characterization of VNA test port cable: transmission phase stability . . . . 87
G.15 Setup for characterization of VNA test port cable: start position . . . . . . . 88
G.16 Setup for characterization of VNA test port cable: end position . . . . . . . . 88
G.17 Characterization of connector repeatability . . . . . . . . . . . . . . . . . . . 90
H.1 Ripple pattern of residual directivity . . . . . . . . . . . . . . . . . . . . . . . 95
H.2 Peak-to-peak magnitude of ripple pattern . . . . . . . . . . . . . . . . . . . . 95
H.3 Ripple pattern of residual source match . . . . . . . . . . . . . . . . . . . . 98
I.1 Directivity ripple in X-band waveguide . . . . . . . . . . . . . . . . . . . . . 106
I.2 Source match ripple in X-band waveguide . . . . . . . . . . . . . . . . . . . 106
J.1 Variations in ripple pattern of residual directivity . . . . . . . . . . . . . . . . 108
J.2 Example uncertainties of Re (S11 ) of a matched load . . . . . . . . . . . . . 110
J.3 Example uncertainties of Im (S11 ) of a matched load . . . . . . . . . . . . . 110
J.4 Example uncertainties of |S11 | of mismatch . . . . . . . . . . . . . . . . . . 112
J.5 Example uncertainties of arg (S11 ) of mismatch . . . . . . . . . . . . . . . . 112

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –6–
J.6 Example uncertainties of |S11 | of short . . . . . . . . . . . . . . . . . . . . . 114
J.7 Example uncertainties of arg (S11 ) of short . . . . . . . . . . . . . . . . . . . 114
J.8 Example uncertainties of Re (S22 ) of an adapter . . . . . . . . . . . . . . . . 116
J.9 Example uncertainties of Im (S22 ) of an adapter . . . . . . . . . . . . . . . . 116
J.10 Example uncertainties of |S12 | of an adapter . . . . . . . . . . . . . . . . . . 119
J.11 Example uncertainties of arg (S12 ) of an adapter . . . . . . . . . . . . . . . 119
J.12 Example uncertainties of Re (S22 ) of a 20 dB attenuation device . . . . . . . 122
J.13 Example uncertainties of Im (S22 ) of an adapter . . . . . . . . . . . . . . . . 122
J.14 Example uncertainties of |S12 | of a 20 dB attenuation device . . . . . . . . . 125
J.15 Example uncertainties of arg (S12 ) of a 20 dB attenuation device . . . . . . 125
J.16 Example uncertainties of Re (S22 ) of a 50 dB attenuation device . . . . . . . 128
J.17 Example uncertainties of Im (S22 ) of an adapter . . . . . . . . . . . . . . . . 128
J.18 Example uncertainties of |S12 | of a 50 dB attenuation device . . . . . . . . . 131
J.19 Example uncertainties of arg (S12 ) of a 50 dB attenuation device . . . . . . 131

List of Tables
8.1 Recommended minimal pin gaps of coaxial connections . . . . . . . . . . . 32
D.1 Error coefficients of one-port error model . . . . . . . . . . . . . . . . . . . . 51
D.2 Error coefficients of ten-term error models . . . . . . . . . . . . . . . . . . . 53
D.3 Error coefficients of seven-term error models . . . . . . . . . . . . . . . . . 55
F.1 Influences in one-port measurement model . . . . . . . . . . . . . . . . . . 68
J.1 Uncertainty budget with rigorous method for Re (S11 ) of matched load . . . 111
J.2 Uncertainty budget with rigorous method for Im (S11 ) of matched load . . . 111
J.3 Uncertainty budget with Ripple Method for S11 of matched load . . . . . . . 111
J.4 Uncertainty budget with rigorous method for |S11 | of mismatch . . . . . . . . 113
J.5 Uncertainty budget with rigorous method for arg (S11 ) of mismatch . . . . . 113
J.6 Uncertainty budget with Ripple Method for S11 of mismatch . . . . . . . . . 113
J.7 Uncertainty budget with rigorous method for |S11 | of a short . . . . . . . . . 115
J.8 Uncertainty budget with rigorous method for arg (S11 ) of a short . . . . . . . 115
J.9 Uncertainty budget with Ripple Method for S11 of a short . . . . . . . . . . . 115
J.10 Uncertainty budget with rigorous method for Re (S22 ) of an adapter . . . . . 117
J.11 Uncertainty budget with rigorous method for Im (S22 ) of an adapter . . . . . 117
J.12 Uncertainty budget with Ripple method for S22 of an adapter . . . . . . . . . 118
J.13 Uncertainty budget with rigorous method for |S12 | of an adapter . . . . . . . 120
J.14 Uncertainty budget with rigorous method for arg (S12 ) of an adapter . . . . . 120
J.15 Uncertainty budget with Ripple method for S12 of an adapter . . . . . . . . . 121
J.16 Uncertainty budget with rigorous method for Re (S22 ) of a 20 dB attenuation
device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
J.17 Uncertainty budget with rigorous method for Im (S22 ) of an adapter . . . . . 123
J.18 Uncertainty budget with Ripple method for S22 of a 20 dB attenuation device 124
J.19 Uncertainty budget with rigorous method for |S12 | of a 20 dB attenuation device 126
J.20 Uncertainty budget with rigorous method for arg (S12 ) of a 20 dB attenuation
device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
J.21 Uncertainty budget with Ripple method for S12 of a 20 dB attenuation device 127

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –7–
J.22 Uncertainty budget with rigorous method for Re (S22 ) of a 50 dB attenuation
device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
J.23 Uncertainty budget with rigorous method for Im (S22 ) of an adapter . . . . . 129
J.24 Uncertainty budget with Ripple method for S22 of a 50 dB attenuation device 130
J.25 Uncertainty budget with rigorous method for |S12 | of a 50 dB attenuation device 132
J.26 Uncertainty budget with rigorous method for arg (S12 ) of a 50 dB attenuation
device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
J.27 Uncertainty budget with Ripple method of S12 of a 50 dB attenuation device 133

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –8–
1 Introduction
1.1 Purpose of this guide
1.1.1 This guide gives advice on how to measure reflection and transmission of guided elec-
tromagnetic waves (scattering parameters or S-parameters) with a vector network an-
alyzer (VNA) in accordance with metrological principles. It is dedicated to advanced
VNA users, i.e. the reader is expected to already have solid knowledge of VNA mea-
surements. For introductory material to vector network analysis the reader is referred
to [1, 2, 3, 4].

1.1.2 While this guide is primarily intended for accredited calibration laboratories it is as well
useful for national metrology institutes and anybody else performing VNA measure-
ments.

1.1.3 This guide explains how to establish SI (international system of units) traceability using
reference standards (section 2). It gives advice on VNA calibration schemes (section 3)
and verification (section 4). It has a particular focus on the evaluation of measurement
uncertainties (sections 5, 6 and 7). Finally, advice on good measurement practice is
given (section 8).

1.1.4 It was the intent to keep the main body of this guide reasonably short with a focus on
practical aspects to make it useful for the practitioner. The appendices contain further
details on many of the topics covered in the main body of the document. Examples
are given to illustrate some of the concepts. A glossary is provided and references are
cited for further reading.

1.2 Comparison with previous guideline


1.2.1 For many years, the previous versions of the guide (version 2.0 and earlier) have served
as a valuable guideline in accredited calibration laboratories and in national metrology
institutes. Advances in VNA metrology over the last years have prepared the foundation
for a revision of the guide. The main improvements are summarized in the following.

1.2.2 The previous versions of the guide promoted SI traceability through beadless air-
dielectric lines using the so-called Ripple Method. This method only partly relies on
a measurement model and makes some questionable assumptions. It has a potential
to either underestimate or overestimate the measurement uncertainty and is limited in
applicability. At higher frequencies it becomes unpractical.

1.2.3 In 2011, Supplement 2 [5] of the ISO-GUM [6] has been published giving thorough
advice on how to evaluate measurement uncertainties for multi-dimensional measur-
ands, including complex-valued quantities. It is therefore the authoritative guideline for
the determination of measurement uncertainties associated with S-parameters. In par-
allel, software capabilities have become available to support the sometimes elaborate
calculation of the measurement uncertainty.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) –9–
1.2.4 This guide therefore promotes the rigorous propagation of uncertainty contributions
through a measurement model that covers the entire measurement process. The re-
fined modeling puts the measured data of the device under test (DUT) in a mathemat-
ical relationship with the characterized values of the calibration standards and other
influence quantities. Thus, it is possible to directly propagate uncertainties associated
with the calibration standards and the influence quantities to the quantity of interest.
This approach takes the multivariate nature (magnitude and phase or real and imagi-
nary components) of the measured quantity fully into account and correlation is treated
properly. Traceability to SI units is established through a set of characterized calibra-
tion standards and assumptions of ideality are not needed. This method is not limited
in applicability and can principally be applied in all areas of VNA metrology. Further-
more, it fits to the purpose of supporting future trends in VNA metrology, such as higher
frequencies, multiport, nonlinear, etc.

1.2.5 This guide gives advice on how to implement the rigorous propagation of uncertainty
to achieve best accuracy in VNA measurements. Due to the widespread use of the
Ripple Method, it is still included in the guide in an improved form, by clearly specifying
its limits in terms of stated uncertainty and applicability.

1.2.6 Compared to previous versions, this guide provides additional advice on VNA mea-
surement models, VNA calibration, VNA verification, and best measurement practice.
Furthermore, it provides an introduction to measurement uncertainties associated with
complex-valued quantities.

1.3 Scope and Applicability


This guide has a focus on linear VNA measurements in the coaxial line system up to
110 GHz. The discussion is limited to one-port and two-port VNA measurements, but
the generalization to an arbitrary number of ports is principally straightforward. Many
aspects can be adopted to other line types such as waveguide or on-wafer. A specific
section on waveguide measurements can be found in appendix I.

1.4 Terminology
Some of the terms used in VNA metrology are rooted in history but are not coherent
with contemporary terminology used in metrology. The subsequent clauses briefly
address these issues.

1.4.1 VNA calibration


In this guide, the term “VNA calibration” is used to describe the determination of the
VNA error coefficients by measuring a set of known reference standards. This termi-
nology is not entirely in agreement with the definition of the term “calibration” in the
International Vocabulary of Metrology (VIM) [7]. The term is nevertheless used here,
because it is widely understood and there are associated terms as “calibration algo-
rithm” or “calibration standard” which would need to be renamed as well. This would
likely lead to confusion.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 10 –
1.4.2 Error model and measurement model
This guide uses the terms “error model” for VNA calibration and VNA error correction
and the term “measurement model” for the evaluation of VNA measurement uncer-
tainties. This distinction is artificial. A VNA error model is nothing other than a VNA
measurement model, reduced to the influencing quantities (error coefficients) that are
determined and corrected during VNA calibration and VNA error correction. The full
VNA measurement model contains all influence quantities, those which can be cor-
rected (the error coefficients) and those which remain uncorrected, as e.g. noise or
drift. It would be more coherent to withdraw the term error model and to just use mea-
surement model. The distinction is nevertheless made, because the term error model
is widely used and replacing it with measurement model might lead to confusion.

1.4.3 Error coefficients and residual errors


Error coefficients are determined during VNA calibration by measuring calibration stan-
dards. Errors in the characterization of these standards lead to errors in the determina-
tion of the error coefficients, something that is commonly referred to as residual errors.
The terminology is not ideal, and it would be better to speak of model coefficients in-
stead of error coefficients (see as well the remarks in 1.4.2). However, due to the wide
spread use of these terms it was decided to keep them in this guide.

2 Traceability schemes and measurement standards


2.1 Traceability chain
2.1.1 S-parameters are dimensionless ratios of complex-valued wave quantities, which are
defined and measured with respect to a reference impedance [8]. The traceability to SI
units is defined through this reference impedance and with respect to a measurement
reference plane.

2.1.2 The first fundamental step of the traceability chain of S-parameters is established with
the characterization of calculable measurement standards [9, 10, 11, 12]. Known cal-
culable coaxial standards are air-dielectric lines, offset shorts, flush shorts and offset
opens. Calculable standards are parametrized and measured dimensionally. The me-
chanical model should cover the whole standard including the connector interface to
achieve highest accuracy and to obtain a consistent definition of the measurement
reference plane [13, 14]. Based on dimensional measurements and known material
parameters the S-parameters of the calculable standards are determined with the help
of analytical equations and numerical EM simulations in combination with electrical
measurements. Calculable standards may be called primary standards. The process
of establishing traceability by using primary standards is called a primary experiment.
Primary experiments are elaborate and require measurement and modeling capabili-
ties at a level that can usually only be provided by national metrology institutes.

2.1.3 A VNA being calibrated by using primary standards can be used to characterize other
calibration standards to disseminate S-parameter traceability to lower levels in the

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 11 –
traceability chain. These standards are called transfer standards (or secondary stan-
dards) and don’t need to be calculable. A good example for transfer standards are
commercially available calibration kits, which contain open, short and matched load
with both male and female connectors.
2.1.4 An accredited calibration laboratory can in principal perform a primary experiment as
well, but due to the required expertise and effort it will normally obtain SI traceability
for its VNA measurements through a set of standards that are calibrated by a national
metrology institute.
2.1.5 VNA parameters, such as power level or frequency, need to be calibrated only if one
of these parameters is important for the characterization of a DUT. An example is the
measurement of a filter, which has a pronounced dependence of transmissivity with re-
spect to frequency. A factory calibration or performance verification (a service typically
offered by service centers of VNA manufacturers) is usually sufficient in these cases.
How often such a factory calibration or performance verification should be performed
can’t be answered generally. It depends on the operating conditions of the VNA. Keep-
ing a history of verification measurements is helpful in this matter, see 4.5.5.

2.2 Measurement standards


Practical advice on use and handling of measurement standards can be found in 8.3
and 8.4.
2.2.1 Two types of measurement standards can be distinguished in VNA measurements,
calibration standards and verification standards. The name calibration and verification
refers to the purpose the standards are used for. Calibration standards are used to cali-
brate a VNA and determine its error coefficients, as discussed in section 3. Depending
on the calibration scheme, the choice of calibration standards may vary. Verification
standards are used to verify that the calibration of the VNA was successful, as dis-
cussed in section 4.
2.2.2 Measurement standards are traditionally mechanical components, which need to be
individually connected to the VNA test port. More recently, electronic calibration units
(ECUs) have become available as a convenient alternative, see 3.3.
2.2.3 Traceability to SI units is established preferably through calibration standards. This re-
quires a model of the measurement process and propagation of uncertainties through
the measurement model, as discussed in 7.1. To establish traceability this way it is
necessary to have SI traceably characterized calibration standards.
2.2.4 Alternatively, by treating the VNA partly as a black box it is possible to establish trace-
ability through beadless air-dielectric lines, as discussed in 7.2. To establish SI trace-
ability this way it is necessary to have SI traceably characterized air-dielectric lines.
With the black box approach it is not necessary to have SI traceably characterized
calibration standards.
2.2.5 Regardless of the method of uncertainty evaluation adopted, it is advisable to hold
an SI traceably characterized set of verification standards. Advice for the selection of

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 12 –
verification standards is given in section 4.

2.2.6 Some of the calibration and verification standards might be of the same type but they
must be physically different items. A single standard must not be used for calibration
and verification simultaneously. The only exception is the re-measurement of the cal-
ibration standards after calibration to test the stability of the measurement system, as
described in 4.3.2.

2.2.7 The memory of older VNAs is limited in size. To cope with this limitation, manufac-
turers of calibration standards used a polynomial representation [3, 15] to model the
frequency response of the highly reflective calibration standards. In addition it is as-
sumed that the matched load is ideal. The measurement accuracy is directly limited
by the quality of the polynomial curve fit, representing the highly reflective standards,
and how well the matched load approximates the reference impedance, e.g. 50 Ω. At
higher frequencies, typically above 2 GHz, a sliding load is used instead of a fixed load
since it provides superior matching performance.

2.2.8 Modern VNAs are able to make use of data-based standard definitions. The S-parameters
of the standards and associated uncertainties are stored as data sets with reasonably
high frequency resolution. Uncertainties associated with the characterization of the
standards determine the measurement accuracy of the calibrated VNA. Measurement
accuracy is not dependent on the fixed load being matched as perfectly as possible.
Thus the use of a sliding load is dispensable. It is still necessary that the standards
behave in a “normal” way, i.e. that the fixed load offers a reasonable matching and the
highly reflective standards principally act as reflecting devices. For best measurement
accuracy, it is essential that the S-parameters of the standards, at any frequency, do
not cluster in a single location of the complex measurement plane.

3 VNA calibration
3.1 Introduction
3.1.1 VNA calibration is the process of determining the VNA systematic measurement errors,
which are called error coefficients (sometimes also referred to as error terms or cali-
bration coefficients). The process is based on a VNA error model. A VNA error model
relates the values indicated by the VNA (raw S-parameters) to the S-parameters of the
device connected to the test port of the VNA. Part of this mathematical relationship are
the unknown error coefficients. During VNA calibration, a set of known calibration stan-
dards is measured and the error coefficients are determined. They can then be used to
transform the raw S-parameters of a DUT measurement into corrected S-parameters,
a step which is usually being referred to as VNA error correction.

3.1.2 VNA error models are based on the idea that the signal flow in a VNA can be modeled
as a linear network. The VNA error model can be illustrated graphically by a signal flow
graph. There are methods for reducing signal flow graphs into mathematical equations,
as e.g. described in [16, 17]. These mathematical equations are used to perform VNA
calibration and VNA error correction.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 13 –
3.1.3 VNA calibration standards are characterized components. It has to be emphasized that
the characterization is never perfect and that the error related to it will be propagated
to the error coefficients and ultimately to the error-corrected S-parameters of the DUT.
It is the purpose of uncertainty evaluation, as discussed in section 7, to estimate the
effect of this influence and other influencing quantities, see section 5.

3.1.4 Since the development of computer-controlled VNAs, several calibration methods have
been developed. The choice between them depends on the final application, the num-
ber and the type of the calibration standards and the desired accuracy. Some details
on one-port and two-port calibration schemes can be found in appendix D.

3.2 Practical advice for VNA calibration


3.2.1 If stable short - open - load (SOL) standards are available and if their characterization
is data-based, the SOL calibration scheme for one-port measurements is preferable.
For two-port measurements, the short - open - load - thru (SOLT) calibration should
be used, if the movement of the test port cable is relatively small and a flush thru
connection is possible (insertable test port configuration). The use of a sliding load is
not advised. See remarks in 2.2.7, 2.2.8 and 8.3.2.

3.2.2 A useful alternative to SOLT is the SOLR (short - open - load - reciprocal, also known
as Unknown Thru) calibration [18]. It uses measurements of short, open and matched
load at each port and the reciprocity condition in the transmission measurement of the
DUT to determine the error coefficients. SOLR should be applied in situations where
the measurement of a flush thru is not practical. This includes cases where the DUT is
of extended size, i.e. a relatively large cable movement would be necessary between
the measurement of the flush thru during calibration and the subsequent measurement
of the DUT. It further includes measurements of non-insertable DUTs and DUTs that
have different connector types and/or different impedance definitions at each port. It
should be noted that applying the SOLR calibration usually requires a four-receiver
architecture of the VNA. Compared to SOLT, SOLR calibration is more sensitive to the
characterization of short and open calibration standards. For further details on VNA
architecture, SOL, SOLT and SOLR refer to appendix D.

3.2.3 Using more calibration standards than needed results in an over-determined calibration
algorithm. For details see D.3. This type of calibration is useful to enhance the accuracy
of the calibration and to detect problems with single calibration standards. The software
(VNA firmware or external program), however, has to to support this type of calibration.

3.3 Electronic calibration units


3.3.1 Electronic calibration units are an attractive alternative to mechanical standards to per-
form VNA calibrations more efficiently. Up to now, ECUs are manufacturer-specific, i.e.
they can only be used with a VNA from the same manufacturer. Remote control of
an ECU from an external software is limited. This can be problematic if it is desired
to propagate the uncertainties associated with the characterization of the ECU states
to the measurement result. If a laboratory intends to establish SI traceability through

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 14 –
an ECU it is recommended to contact the manufacturer and/or a national metrology
institute for advice. Stability effects of ECUs related to temperature and aging are
addressed in 8.4.

4 Verification
4.1 Introduction
According to the VIM [7], verification generally denotes provision of objective evidence
that a given item fulfills specified requirements. In the present case of VNA measure-
ments, the purpose of verification is to confirm that the VNA, test port cables, and
calibration standards are working together correctly as a measurement system, i.e.,
that the stated measurement uncertainty is met. This section has a focus on practical
advice related to verification. More details are provided in appendix E.

4.2 Purpose of verification


After a VNA calibration has been completed, it is necessary to investigate the valid-
ity and quality of the calibrated measurement setup. The measurement uncertainty is
calculated from the influences discussed in section 5. Underestimation of these influ-
ences or any additional source of error, as e.g. faulty calibration caused by mixed-up,
damaged or unstable calibration standards, use of the wrong calibration standard defi-
nitions, incorrect operation, non-linearity due to too excessive source power, VNA fail-
ure and a faulty setup (e.g. a loose connection) might not be covered by the calculated
measurement uncertainty. It is the purpose of verification to discover such sources of
error.

4.3 Verification Methods


Verification can be performed in various ways. Besides quantitative coincidence tests,
see 4.3.1 and E.2, it is also possible to perform more qualitative plausibility tests, see
4.3.2, that might be helpful to analyze potential problems with the measurements.

4.3.1 Coincidence tests


Verification is generally performed by measuring characterised, stable, high-accuracy
standards and performing a coincidence test using their reference data, e.g. from a
calibration certificate. Different verification standards and their response to the residual
errors of the VNA are discussed in E.1. Fail/pass criteria for coincidence tests are
discussed in 4.5 and E.2.

4.3.2 Plausibility tests


Plausibility tests address the soundness of the VNA calibration in a more qualitative
way by investigating frequency response and physical limits of standards, based on
fundamental considerations.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 15 –
4.3.2.1 An initial test after calibration can be performed by re-measuring the calibration stan-
dards. If the re-measurement does not precisely match the characterization data of
the calibration standard, there is a stability or noise issue in the measurement system.
Note that the re-measurement of calibration standards is not a verification of the cali-
bration. It neither identifies faulty calibration, by using the wrong calibration standard
definitions or mixed up standards, nor potential malfunction of the standards. A set of
standards that is physically different from the calibration standards needs to be used
for that purpose, see 2.2.6.

4.3.2.2 In the most common case of the reflection coefficient S11 of mechanical offset opens
and offset shorts, the frequency response loci should produce approximately circular
arcs on the Smith Chart edge approaching the reflection coefficient S11 = 1 point
(right-hand side) for the opens and S11 = −1 (left-hand side) for the shorts for low
(MHz) frequencies, respectively.

4.3.2.3 The measurement of a flush short is suitable to test how well the reference plane has
been defined by VNA calibration. As a flush short is a passive device it is expected
that the magnitude of the reflection coefficient will not exceed |S11 | = 1. A perfect
flush short would show no phase change with frequency. Due to imperfections in the
fabrication process this is, however, never the case. In reality the phase should show
a slow monotonous clockwise change starting at -180 deg. Any deviation from this
behavior is an indication that the reference plane has not been properly defined during
VNA calibration, see e.g. [13].

4.3.2.4 Applying different VNA calibration algorithms, preferably with different sets of calibra-
tion standards, should generally lead to equivalent results for the measurements of
the DUTs, conditional to the associated measurement uncertainties. An effective and
also quantitative test is the comparison of SOLT and SOLR calibration, as discussed
in E.1.2.6.

4.3.2.5 After an SOLT calibration it is useful to measure a reciprocal two-port device with finite
electrical length to verify the symmetry, i.e. S21 = S12 . Any significant deviation would
indicate problems related to the calibration.

4.3.2.6 If the frequency range is divided up into bands, and different sets of calibration stan-
dards or different calibration methods are used, then measurement results should show
fairly continuous transitions when crossing between frequency bands.

4.4 Verification procedure and practical advice


4.4.1 In a strict sense, complete verification includes all S-parameters to be tested at each
measurement frequency point. Since in case of data-based calibrated verification stan-
dards, reference data are available only for a limited number of frequency points, and
furthermore, since there is certain correlation between nearby frequency points, the
existing frequency grid will sometimes not be in line with the demands of the DUT. How-
ever, if reference data for the verification standards are available it is recommended to
choose this frequency grid for VNA calibration and DUT measurement too.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 16 –
4.4.2 The uncertainty associated with the characterization of the verification artifacts should
be smaller (or at least not larger) than the uncertainty associated with the measure-
ments of the verification artifacts.

4.4.3 Verification and calibration standards that have been calibrated by the same laboratory
can potentially be used to apply sharper verification criteria, see 4.5.6. It is therefore
recommended but not necessary.

4.4.4 As a general rule it is advised to perform a coincidence test using at least one verifica-
tion standard having a response similar to the DUT.

4.4.5 For a more general verification covering the entire region of the Smith Chart the verifi-
cation should be performed as a two-step procedure. First, the one-port performance
of all ports has to be checked by measuring calibrated one-port verification standards.
This is followed by the two-port verification.

4.4.6 For the one-port verification it is recommended to measure two highly reflective stan-
dards with different phase (different offset lengths compared to calibration standards)
and one matched load. The matched load might be replaced by a matched attenuation
device with ≥30 dB of attenuation. To check the non-linearity of the measurements it
is also recommended to measure a mismatch standard with a reflection coefficient of
approximately 0.2 to 0.5.

4.4.7 For the two-port verification a transparent device (beaded air-dielectric line or adapter)
and a set of matched attenuation devices should be measured. Most manufacturer
verification kits contain a 20 dB and a 40 dB attenuation device. It is recommended to
perform an additional measurement using a 3 dB attenuation device to have a better
coverage of the Smith Chart and to detect problems related to non-linearity.

4.5 Verification criteria


4.5.1 Pass/fail criteria for verification measurements have to take into account that the mea-
surement and the reference data of the verification device each have an associated
uncertainty. A statement about the agreement between measurement and reference
data is therefore always a probability statement. The acceptable probability level is
usually determined by common sense and by convention.

4.5.2 Generally, the result of a linear VNA measurement is expressed as frequency-dependent,


complex-valued S-parameters. In most cases, characterization data along with its as-
sociated uncertainties are given in magnitude and phase or alternatively, in real and
imaginary part. The uncertainty (covariance) matrix might be known as well. Verifi-
cation can be performed on a single scalar parameter, e.g. magnitude or phase or on
the full two-dimensional complex-valued quantity. The former is justified if there is only
interest in a single scalar component of the DUT’s S-parameters.

4.5.3 Visual inspection of the frequency response of the data, i.e. the difference between
measurement and characterization data with associated uncertainties, is highly rec-
ommended. It allows a judgement to be made about the quality of the verification

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 17 –
and helps to identify problems. In most cases visual inspection is sufficient to identify
potential problems.

4.5.4 Quantitative evaluation of verification data requires mathematical analysis of the data
and the associated uncertainties. Automation of data analysis might be useful if a large
amount of data has to be evaluated. A method for quantitative evaluation of verification
data is shown in E.2.

4.5.5 In either case, visual or quantitative evaluation, it is recommended to build a history


of the measurements of the verification standards. This is helpful to detect gradual
changes, e.g. if a calibration standard is drifting (long-term stability). The first mea-
surement of the verification standard is taken as the reference. Subsequent measure-
ments give an indication of the size of random effects affecting the measurement. Any
deviation beyond that or a systematic trend indicates a problem.

4.5.6 If calibration standards and verification standards have been calibrated by the same
laboratory, there is likely to be a strong positive correlation between standards of the
same type, e.g. the matched load used for calibration and the matched load used for
verification. Consequently this will lead to a similar correlation between measurement
and reference data of the verification standard. This will further lead to a smaller un-
certainty associated with the difference between measurement and reference data and
therefore to a sharper pass/fail decision. However, this can only be exploited, if the
correlation information between calibration and verification standards is available and
if the evaluation of the uncertainty associated with the measurement of the verification
standards considers this correlation information to the full extent. Obviously, quanti-
tative evaluation according to the formalism in E.2 is needed to take correlation into
account.

5 Uncertainty contributions
5.1 Introduction
This section identifies and discusses quantities influencing VNA measurements. There
are a number of quantities with significant influence on the result of a VNA measure-
ment. The measurement models discussed in section 6 refer to these quantities ex-
plicitly. Estimates of these quantities, and the uncertainties associated, are inputs to
the measurement models to determine a value and uncertainty associated with the
quantity of interest, usually the S-parameters of the DUT.

5.2 Identification of influence quantities


5.2.1 The accuracy of S-parameter measurements is affected by the following influence
quantities:

• Characterization of calibration standards

• VNA noise floor and trace noise

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 18 –
• VNA non-linearity

• VNA drift

• Isolation (cross-talk)

• Test port cable stability

• Connection repeatability

In some specific cases other VNA parameters such as frequency or output power have
to be considered in the uncertainty analysis too. However, this is rather unusual and
therefore not considered here (see also 2.1.5).

5.2.2 All these influences are sources of measurement errors. It is therefore essential that
they be part of any model that describes the measurement and that is being used to
evaluate the uncertainty of the measurement. Any uncertainty evaluation that does not
address all the components listed above has to be considered as incomplete.

5.3 Characterization of uncertainty contributions


Below, the influences listed in 5.2.1 are discussed. Some of them depend on the set-
ting of the VNA, measurement setup, environmental and operating conditions, etc. The
characterization procedures to determine uncertainty contributions based on these in-
fluences are not unique and strongly depend on the measurement model being used.
Detailed step by step procedures are shown in appendix G and are specifically dedi-
cated to the measurement models in appendix F.

5.3.1 Characterization of calibration standards

5.3.1.1 Characterized calibration standards are necessary if the uncertainty evaluation is per-
formed with the rigorous method, according to 7.1. If the Ripple Method is applied,
according to 7.2, it is also possible just to use manufacturer data, which is often pro-
vided along with the calibration kit.

5.3.1.2 The reference data of calibration standards with associated uncertainties are taken
from a calibration certificate, unless the laboratory has its own realization of traceability
to SI units. Normally, this is only the case for national metrology institutes.

5.3.1.3 Usually the dominant source of uncertainty in a VNA measurement are associated with
characterisations of calibration standards. Accurate characterization of the calibration
standards includes both body and connector [13]. For an accurate determination of
the measurement uncertainty it is preferable that correlations between the uncertainty
components of the calibration standards are taken into account. Neglecting correla-
tions can lead to either underestimation or overestimation of the measurement uncer-
tainty.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 19 –
5.3.1.4 Older VNAs require the calibration standards to be characterized by polynomial coef-
ficients (see 2.2.7 as well). A polynomial representation was used in the past due to
the memory limitations of the VNA firmware. Modern VNAs allow characterization data
to be stored in a data-based form, i.e. as a frequency dependent list of reference val-
ues. This is generally a more flexible and accurate way to store characterization data.
The data-based characterization can also be used with older VNAs if an appropriate
software, e.g. [19], is applied to perform calculations externally on a computer.

5.3.1.5 Calibration standards can drift or change unexpectedly over time. A slow drift due to
aging is normal and it is therefore necessary to recalibrate the standards on a regular
basis. The recalibration interval depends on the operating conditions and needs to be
defined on a case to case basis. Unusual drift or change can be detected through
verification, see section 4. Good care and best measurement practice, see section 8,
are advised to minimize aging. Aprupt changes or drift in the calibration standards can
also be detected by using more calibration standards than necessary and performing
an over-determined calibration, see D.3.

5.3.2 Noise floor and trace noise


An example of a step by step characterization of noise floor and trace noise is given in
G.1.

5.3.2.1 Noise denotes random signal fluctuations, which are characteristic of all electronic
circuits. For VNA measurements two types of noise are distinguished. Noise floor
denotes random fluctuations in the absence of a deterministic signal. Trace noise de-
notes random fluctuations of the measurement result. When trace noise is far above
the noise floor, it is proportional to the measured value, i.e., it is a constant fraction
of the result. In absolute notation, however, trace noise grows in proportion to an in-
creasing measurement result. The noise content of a transmission coefficient can thus
be characterized by a single dB value, as far as the contribution from the noise floor
is negligible. Noise floor and trace noise are dependent on the setting of the VNA, for
additional information see 8.2.5 and 8.2.6. Smaller IF bandwidth and averaging reduce
the effect.

5.3.2.2 Noise contributions to the measurement uncertainty are in principle already included in
the characterization of the connection repeatability, see 5.3.7. It is not possible to sep-
arate the two effects and one might be tempted to neglect an extra noise contribution
to the measurement uncertainty. Neglecting an extra noise contribution is principally
possible but care should be taken, because the noise contribution varies, depending
on the VNA settings and also depending on the S-parameters of the DUT.

5.3.3 VNA non-linearity


An example of a step by step characterization of VNA non-linearity is given in G.2.

5.3.3.1 The VNA is being modeled as a linear network, i.e. there is a linear relation between
the forward and backward propagating waves and their detection by the receivers. The
term VNA non-linearity, sometimes also called dynamic accuracy, denotes deviations
from this behavior. The effect is dependent on the setting of the source power level

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 20 –
of the VNA. The frequency dependence is generally weak, unless compression effects
are present. Compression effects should be avoided by choosing appropriate settings
of the VNA, see 8.2.
5.3.3.2 VNA non-linearity combines effects from different components: amplifiers, filters, ADC
non-linearities, etc. Noise also results in an apparent non-linear behavior at low signal
levels, but this effect is already addressed in 5.3.2. Refined modeling could assign
individual uncertainties to each of the contributors. A review of existing methods to
determine the non-linearity of receivers by means of step attenuation devices or power
sensors as well as a method based on two phase-locked signal sources is given in [20].
5.3.3.3 Due to the lack of access to the individual components in a VNA it is usually more prac-
tical to characterize non-linearity at the level of receiver signal ratios, i.e. S-parameters.
The step by step characterization in G.2 determines the non-linearity of receiver signal
ratios with the help of a step attenuation device.
5.3.3.4 The characterization of step attenuation devices over a wide frequency range may be
quite laborious. Alternatively, the procedure described in G.2 is only applied at low
frequencies, whereas the non-linearity of the upper frequency range is checked with a
characterized fixed attenuation device, e.g. 10 dB.
5.3.3.5 The characterization in G.2 aims at determining residual non-linearities with the mea-
surement of passive devices, assuming that the source power level has already been
reduced such that compression effects are avoided. Further information on how to
ensure this is given in 8.2.
5.3.4 VNA drift
An example of a step by step characterization of VNA drift is given in G.3.
5.3.4.1 Drift in VNA measurements occurs due to temperature changes and resulting relax-
ation effects. These effects lead to changes in the electrical length of signal paths and
to changes in the performance of couplers, receivers, and other components. Some
of the dielectric materials used in test port cables are phase sensitive to temperature
changes. At the measurement level this leads to changes in the error coefficients that
have been determined during the calibration of the VNA. Drift is not just a phenomenon
of the VNA itself, but strongly dependent on environmental factors, measurement setup
and operating conditions. Furthermore, drift effects are dependent on the time elapsed
after the calibration of the VNA.
5.3.4.2 Drift can be accounted for in different ways. With refined modeling it is in principle
possible to assign drift parameters to individual components of the VNA, possibly taking
temperature readings into account. In practice it might be more feasible to model drift
at the level of the linear network, as shown for the measurement models in appendix F.
Drift is assigned to each individual error coefficient. Procedures are described in G.3.
5.3.4.3 A simple and practical procedure to determine drift effects is to measure the calibra-
tion standards twice, at the beginning and at the end of a measurement session, and
subsequently evaluate the differences between both data sets in terms of drift terms
for the individual error coefficients.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 21 –
5.3.5 Isolation (cross-talk)

5.3.5.1 Isolation is represented by error coefficients in VNA error models, e.g. E03 and E30 in
figure D.5. It is generally of minor importance in coaxial measurements with modern
VNAs. For older VNAs it is known that it affects measurements of large attenuation
to the extent that a correction needs to be applied. Because the effect might as well
depend on the setup and has a potential to drift it should be characterized with each
measurement, as it is done for the other error coefficients.

5.3.5.2 The treatment of isolation, as done in appendices D and F, might be inadequate for
measurements other than coaxial. E.g., for on-wafer measurements additional leakage
and cross-talk paths need to be considered. This is not further discussed here. More
information can be found in [21, 22].

5.3.5.3 The procedure of characterization follows the characterization of the noise floor, as
described in G.1. The uncertainty associated with the isolation terms cannot be sepa-
rated from the noise floor. It is therefore usually justified to assume that the uncertainty
is already included in the noise floor.

5.3.5.4 If the characterization G.1 results in a mean of S21 or S12 that is significantly different
from 0, e.g. if the difference is larger than the standard uncertainty associated with the
repeated measurements of S21 or S12 , the measurement of the DUT needs to be error
corrected using the mean value of S21 or S12 as the error coefficient related to isolation.

5.3.6 Test port cable stability


An example of a step by step characterization of test port cable stability is given in G.4.

5.3.6.1 Test port cables are sensitive to temperature changes, movement and other mechan-
ical influences. This section addresses the sometimes unavoidable movement of test
port cables. Temperature effects of cables are summarized under drift effects, as dis-
cussed in 5.3.4. Other influences inducing mechanical stress should be avoided as far
as possible, see 8.5.

5.3.6.2 Moving the test port cable(s) during calibration or DUT measurement will change the
error coefficients of the VNA. Thus, cable movement and cable torsion should be
avoided as much as possible. Related best practice is discussed in 8.5.1. For measure-
ments that involve more than one VNA port, cable movement is inevitable. The effects
of cable movement are strongly dependent on the quality of the test port cable. Even
for cables of the same type differences in stability might be rather large. Furthermore,
there is a dependency on the measurement setup and how the cables are connected to
the VNA, see 8.5. Therefore it is recommended to individually characterize the cables
in the specific measurement setup.

5.3.7 Connection repeatability


An example of a step by step characterization of connector repeatability is given in G.5.

5.3.7.1 Deviations from ideal connector geometry lead to mechanical stress and unwanted
deformations when mating connectors. This causes changes in electrical behavior for

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 22 –
re-connections under different connector orientations. This effect is dependent on each
individual combination of connector pairs. Nevertheless, it is possible to determine a
typical repeatability for different connector families as described in example in G.5.
5.3.7.2 It is, however, still recommended to measure at least four re-connections under different
azimutal positions for all components that are connected to the test port to verify the
typical performance, see 8.8 for further discussion.
5.3.7.3 Furthermore, for a reliable characterization of repeatability it is necessary to use test
port adapters with recessed pin depths, as described in 8.5.2, to avoid near-field cou-
pling and potential resonant behavior [23].

6 Measurement model
6.1 A VNA measurement model is the basis for the evaluation of the measurement un-
certainty. The measurement model is an analytical expression that relates inputs and
influence quantities to the final output quantities, as discussed in more detail in C.3. In
VNA metrology it relates the influence quantities discussed in 5.3 to the S-parameters
of the DUT. Using methods of uncertainty propagation, as discussed in section 7, the
uncertainties associated with the influence quantities can be propagated to evaluate
the uncertainty associated with the final result.
6.2 A measurement model is not necessarily a single large equation. It is often more
convenient and more natural to have a set of connected equations, each of which
representing either a different part of the measurement process, e.g. VNA calibration
and VNA error correction, or a refined model of one of the influences, e.g. a drift model
taking time spans between measurements and/or temperature readings into account.
6.3 A measurement model of the VNA is closely related to the error model that is applied
during VNA calibration. See appendix F for details.

7 Uncertainty evaluation
Two methods of uncertainty evaluation are presented below, the rigorous method with
uncertainty propagation through a full measurement model and the Ripple Method. It
should be pointed out that both methods deliver the same measurement result (i.e., the
same estimates of S-parameters of the DUT), when the same standards and calibration
scheme are used for VNA calibration. The different methods will, however, obtain
different measurement uncertainties for the same result. For general remarks on S-
parameter uncertainties refer to appendix C. The result of any uncertainty evaluation
should be verified regardless of the applied method, see section 4 and appendix E.

7.1 Rigorous uncertainty evaluation


7.1.1 Rigorous uncertainty evaluation follows the methodology recommended by the ISO-
GUM documents by using a full measurement model and taking the multivariate char-
acteristics of the quantities fully into account. It is not limited in applicability and it is also

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 23 –
suited to be applied at higher frequencies. It requires, however, the use of specialized
software to carry out the elaborate calculations.

7.1.2 Rigorous uncertainty evaluation is based on a measurement model, which covers the
entire measurement process, i.e. the calibration of the VNA using calibration stan-
dards and subsequent error correction of a measurement of the DUT. Details on the
measurement model are provided in section 6 and appendix F. An essential part is
the characterization of the basic influence quantities. Following the instructions in 5.3
measurement uncertainties are assigned to the basic influence quantities. These un-
certainties are propagated through the measurement model by linear or numerical (e.g.
using the Monte Carlo Method in [5]) uncertainty propagation.

7.1.3 Based on the discussion in C.4 it might be more economical to use linear uncertainty
propagation instead of numerical methods. However, the equations of linear uncer-
tainty propagation, taking correlations fully into account, are elaborate, see [24]. It is
beyond the scope of this guide to write these equations down, because the evaluation
by hand or in a spreadsheet is not feasible. Instead software support is needed.

7.1.4 Suitable software solutions, which are able to handle the uncertainty propagation of
complex-valued quantities, are available [25, 26, 27]. These tools provide general
frameworks to realize custom-built implementations of rigorous S-parameter uncer-
tainty evaluation. For software solutions that are specifically targeting S-parameter
measurements see [28, 19, 29, 30]. These solutions already contain the VNA mea-
surement models and support different calibration algorithms, i.e. programming is not
necessary. Questions related to validation of a software are not addressed in this
guide, because they can’t be generally answered and have to be evaluated case by
case.

7.1.5 Depending on the implementation of rigorous uncertainty propagation it not only pro-
vides the correlation between the two components of a complex-valued S-parameter,
but also the correlation information between the individual S-parameters and possibly
even the cross-frequency correlations. This is valuable if the measured S-parameters
are used to calculate derived quantities.

7.1.6 It is a disadvantage of the rigorous method that the equations of the measurement
model can’t be easily printed in this guide and the user needs to rely on external soft-
ware. All other elements that are needed to implement the method are available in this
guide.

7.1.7 The traceability to SI units is established through the calibration standards. It is there-
fore necessary to have an SI traceably characterized calibration kit to implement the
rigorous method. See the remarks in section 2 as well.

7.2 Ripple Method


7.2.1 The Ripple Method has a long history of usage. It was developed as a pragmatic
and adequate approach to measurement uncertainty evaluation when data processing

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 24 –
technologies, computational power and software solutions were not as advanced. The
method has already been described in the previous editions of this guide.
7.2.2 The Ripple Method starts with the calibrated VNA. It uses measurements with a bead-
less air-dielectric line to determine the residuals of the error coefficients. These resid-
uals are used to assign measurement uncertainties to the error coefficients. These
uncertainties are propagated through a reduced measurement model, the so-called
residual model (F.5), to assign a measurement uncertainty to the DUT measurement.
7.2.3 The Ripple Method has shortcomings and disadvantages, some of which are inherent
to the method (at least if the method is applied in its simplest form) and others that are
related to the implementation in the previous versions of this guide.
The method requires the handling of beadless air-dielectric lines, which becomes more
tedious with increasing frequencies. The air-dielectric lines are assumed to be ideal,
neglecting the reflection at the connector interfaces and any losses [31]. The method
is unable to determine the residual error of the tracking term. Furthermore, it is limited
to the magnitudes of the residuals and the measurand. It does not determine an uncer-
tainty associated with the phase. The method is unable to take correlation effects into
account. Uncertainty information is limited to discrete frequencies, the spacing of which
is determined by the length of the air-dielectric line. Abrupt changes with frequency in
the residual error coefficients can be problematic. The residual measurement model
assumes that the VNA is measuring error corrected S-parameters, hence influences of
raw error coefficients are neglected.
The implementation of the Ripple Method in the previous versions of the guide lacked
a measurement model considering all the influences listed in section 5. Important con-
tributions to the measurement uncertainty, such as e.g. cable movements in two-port
measurements, have been dealt with improperly. See as well [32] for an evaluation
of the equations used in the previous versions of the guide with respect to form and
content.
7.2.4 The Ripple Method is retained in this guide for backward compatibility, acknowledging
the fact that many laboratories are still applying this method. The method is presented
here in a modified form to eliminate some of the shortcomings mentioned in 7.2.3.
Details of the implementation are described in appendix H.
7.2.5 Due to the reflections at the connector of the air-dielectric line it is possible to obtain
large variations of the ripple amplitude depending on the position of the center conduc-
tor. Together with the inability to take correlation effects into account this can potentially
lead to a significant overestimation or underestimation of the measurement uncertainty.
The treatment in appendix H has therefore built in some safeguards to avoid underes-
timation. As a consequence the uncertainties that are obtained by the Ripple Method
are in some cases significantly larger when compared to rigorous uncertainty propa-
gation, as can be seen in the examples in appendix J.
7.2.6 It should be pointed out that the worst case measures that are taken in appendix H aim
solely to protect from unrealistic small measurement uncertainties. If other measures
are taken, as e.g. those mentioned in 7.2.8, it is possible to adjust the evaluation ac-
cordingly and relax some of the worst case restrictions. This needs to be decided on a

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 25 –
case by case basis.

7.2.7 An advantage of the Ripple Method is that everything needed to apply the method is
documented in this guide and it is not necessary to rely on external specialized software
to implement it. The equations in appendix H are sufficiently simple to be implemented
in a spreadsheet.

7.2.8 It has been demonstrated that it is possible to increase the quality of the Ripple Method
with considerable additional effort. Many of the shortcomings mentioned in 7.2.3 can
be resolved this way. The enhanced method requires a characterized air-dielectric line
with a positionally controlled center conductor and advanced methods of data process-
ing. Due to the complexity of the method it is not considered in this guide. Though
it should be pointed out that in this improved form the Ripple Method can be a valid
alternative to evaluate the measurement uncertainty of S-parameter measurements.
For details see [33, 34, 35, 36, 37, 38].

8 Best measurement practice and practical advice


Measurement models are only approximations of reality. Influences that are not in-
cluded in the model will lead to degraded results and erroneous measurement un-
certainties. VNA measurements are demanding and susceptible to mistakes. Good
measurement practice is therefore an inevitable requirement for performing high qual-
ity measurements with associated uncertainties that are reliable.

8.1 Environmental conditions


8.1.1 The standard laboratory temperature for electrical measurements is 23 ◦ C. Deviations
from the standard temperature might change the characteristics of the calibration stan-
dards and introduce an additional error during VNA calibration. This effect depends on
design details of the calibration standards. User guides of calibration kits often specify
environmental requirements. As a rule of thumb deviations up to ±1 ◦ C from stan-
dard temperature are of no concern for the characteristics of typical VNA calibration
standards.

8.1.2 Changes in the temperature during and after calibration might lead to drift in the er-
ror coefficients of the VNA and will directly affect measurements of the DUT. This is
generally more significant than the effect of absolute deviation from standard labora-
tory temperature on the calibration standards. Drift occurring after calibration can be
specified for specific environmental conditions, see 5.3.4, and needs to be considered
in the evaluation of the measurement uncertainty. A measurement setup including test
port cables is more sensitive to temperature effects. On the other hand a test port
cable decouples the measurement devices from the VNA test port. Heat flow from the
VNA test port might otherwise lead to heating of the components and change their
electrical characteristics. For reasonably stable environmental conditions it is therefore
recommended to use a test port cable.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 26 –
8.1.3 Special care should be taken to avoid local heating through unnecessary touching of
components and connectors. The effect can be minimized through the use of gloves
or finger cots. Adequate waiting time between manipulation and measurement should
be considered for cool off.
8.1.4 After an extended period of shutdown VNAs should be allowed to reach thermal equi-
librium before resuming operation. If the VNA has been exposed to extreme temper-
atures, i.e. during a transport in winter time, stabilization might take several hours or
even days.
8.1.5 Humidity has low impact on VNA measurements as long as extreme values are avoided.
A relative humidity of (40 ± 20)% is recommended. High humidity might cause un-
wanted effects due to condensation. Very low humidity increases the chance of dam-
aging the VNA due to electrostatic discharge.
8.1.6 Operating the VNA in a rack cabinet with appropriate ventilation and proper manage-
ment of the power supply is recommended. It results in more stable temperature con-
ditions.

8.2 VNA architecture, performance and settings


8.2.1 The four-receiver architecture of VNAs is advantageous for accurate measurements,
because it allows to directly measure the VNAs switch terms. Furthermore, calibration
algorithms other than SOL and SOLT can be applied. For more details see D.2
8.2.2 Not all VNA errors caused by hardware imperfections are corrected during calibration.
An example is non-linearity. Constructional specifics of a VNA might influence its drift
behavior. VNAs with good drift stability are less sensitive to environmental changes
and help to increase the accuracy of the measurements.
8.2.3 The error coefficients source match and load match, see D.2.2, are corrected through
calibration. However, it is better if the VNA has a good raw match performance, be-
cause both, source match and load match, have a tendency to magnify the measure-
ment errors that arise when characterising highly reflective calibration standards. De-
pending on the calibration scheme, this enhancing effect can be more or less pro-
nounced.
8.2.4 Compression effects can affect both reflection and transmission measurements. The
source power of the VNA should be chosen so that compression effects are avoided.
This can be checked using an uncalibrated fixed attenuation device (3 dB or 10 dB),
which is measured repeatedly with the error corrected VNA, starting at a high source
power and then gradually decreasing it. Stabilization of the measured attenuation value
over the whole frequency range indicates the maximum permissible source power for
compression-free operation. For reflection measurements the same procedure can be
carried out with a mismatch standard.
8.2.5 The impact of noise on measurement can be lowered by reducing the IFBW (interme-
diate frequency bandwidth) and/or applying averaging. Both approaches will increase
the measurement time, though IFBW does so to a lesser extent. The impact of drift

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 27 –
scales with the measurement time. Therefore, as a rule, averaging should only be ap-
plied if further reduction of the IFBW does not result in an improvement. There are two
basic types of averaging available, point by point averaging (multiple measurements
at each frequency point) and trace averaging (multiple measurements of the whole
trace). Contemporary VNAs usually support both types. The two methods differ in
how drift becomes noticeable in the measurement results. There is no general prefer-
ence between point and trace averaging and the choice depends on application and
measurement conditions.

8.2.6 Noise can be reduced through smoothing as well. This functionality should be used
with care, however. Pronounced features in the frequency dependence, e.g. in a mea-
surement of a filter, might be altered. Smoothing has the potential to introduce addi-
tional measurement error if used negligently.

8.3 Use of mechanical calibration and verification standards


8.3.1 Mechanical calibration standards are usually taken from a commercial calibration kit. It
is not advised to mix open and short standards, respectively, originating from different
calibration kits, unless they are matched in their electrical lengths.

8.3.2 Many calibration kits contain sliding loads to replace the fixed load at higher frequen-
cies. A sliding load increases the accuracy of the measurement if the physically im-
perfect fixed load is assumed to be ideal, i.e. without reflection. The assumption of an
ideal load is usually made in combination with the polynomial representation of open
and short. A sliding load is not needed if the definition of the fixed load is data-based.
See comments in 2.2.7 and 2.2.8.

8.3.3 A key quality characteristic of mechanical standards is stability, which has a direct
impact on the accuracy of measurement. A high connection repeatability with low
sensitivity to connector orientation is essential (see 5.3.7 and 8.8), as is long term
stability (see 5.3.1.5).

8.3.4 Quality criteria for mechanical calibration standards that can be assessed visually are:
concentricity of center conductor, integrity of contact fingers of female interfaces and
surface finish of the electrical contact areas. The electrical stability of the standards is
directly impacted by these factors.

8.3.5 Some mechanical calibration and verification standards or test port adapters are avail-
able in two different designs of the female connector interface, slotted and slotless.
Slotted interfaces are more sensitive to variations of the male pin diameter compared
to an optimized slotless design. Slotless designs, on the other hand, might have a
larger reflection coefficient. From a metrological point of view the latter is of less con-
cern. A slotless design is therefore preferable, because it generally provides better
repeatability. See the notes in 8.5.2 as well.

8.3.6 Calibration standards can change their characteristics because of heat flow caused by
mating with a warm test port or by touching, see 8.1. Both should be avoided as much
as possible. Typical indicators of this problem are phase changes of highly reflective

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 28 –
components or changes in the magnitude of the reflection coefficient of matched com-
ponents at low frequencies.

8.4 Use of electronic calibration units


8.4.1 The impedance states of ECUs are susceptible to environmental temperature changes.
Therefore in most cases ECUs have an internal heater to keep the impedance stan-
dards at a constant temperature. This temperature is usually several degrees above
the standard laboratory temperature. Thus, heat flow occurs from the ECU, as soon
as the ECU is connected to the VNA test port or the test port cable. The achievable
measurement accuracy is therefore limited, because the heat flow changes the error
coefficients of the VNA [39].

8.4.2 The impedance states of an ECU are subject to aging and hence the recalibration
intervals have to be chosen accordingly. Long-term stability tests under controlled
laboratory conditions are indicating that ECU states are stable on the time scale of
months and years. However, this is very much dependent on the design of the ECU
and on the operating conditions. It is therefore impossible to generally quantify the
recalibration interval. It is advised to start with short recalibration intervals to build an
evidence-based history of the ECU and then adjust the interval accordingly.

8.5 Test port cables


Apart from non-linearity and noise effects, the stability of test port cables is often the
most important uncertainty contributor for transmission coefficient measurements. Us-
ing high performance test port cables is generally beneficial in terms of measurement
uncertainties. An appropriate fixture and layout, see 8.5.1, helps to minimize cable
effects. The SOLR calibration scheme, see D.2.3, is an alternative to SOLT that can
avoid excessive cable movement. The procedure to evaluate the stability of cables is
described in G.4. Appropriate test port adapters, see 8.5.2, act as test port protectors
and avoid near field resonance effects.

8.5.1 Fixture and layout

8.5.1.1 An appropriate fixture that supports and defines the layout of the test port cables is a
necessary accessory for accurate VNA measurements. A simple cable fixture can be
realized with a plain and solid surface in front of the VNA (e.g. a commercially available
breadboard) using foam pads and clamps. An example is shown in figure 8.1. The
fixture should keep the test port cable fixed during one-port measurements. For two-
port measurements one cable remains fixed and the other one should be arranged so
that unnecessary cable movements are avoided. If possible, the test port cables should
be maintained in a horizontal plane defined by the test port connectors of the VNA.
Cable twisting, extreme curvature and stress on the connection should be avoided.

8.5.1.2 High performance test port cables often have a natural bend due to the production pro-
cess. Maintaining the natural bending helps to reduce measurement error due to cable
movement. Even then it is still possible that small angular changes in the connection
of the test port cable to the VNA can lead to differences in phase stability. This can

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 29 –
Figure 8.1: Recommended setup of VNA test port cables. A breadboard serves as work
space. The cables are supported by foam pads to avoid sagging from the VNA test ports.
Clamps are used to keep the cables fixed during measurements.

be examined and optimized by connecting a short at the reference plane (end of the
test port cable) and observing how phase variations depend on cable movement while
changing slowly the angular orientation of the connection between test port cable and
VNA.

8.5.1.3 For reasons given in 8.5.2 it is recommended to use an adapter in combination with
the test port cable. The most stable cable-adapter combination is achieved by using
ruggedized connectors at the cable-adapter interface. The ruggedized connector can
be found on many VNA test ports. It has a large threaded body that helps to stabilize
the connection.

8.5.1.4 The selection of the correct VNA calibration scheme can help to reduce measurement
error due to cable movement as well, as discussed in 3.2.

8.5.2 Test port adapters

8.5.2.1 It is recommended to use test port adapters for several reasons. If any damage occurs
to the connector only the adapter has to be changed or repaired, rather than the entire
test port cable or VNA test port. Using a metrology grade adapter with geometries
close to the nominal values and good concentricity of the center conductor will minimize
mechanical stress and improve electrical repeatability. Finally, an optimal pin depth with
respect to the reference plane, see 8.5.2.2, can be established on the test port side.

8.5.2.2 In the past, the objective was always to have a connection at the measurement refer-
ence plane as flush as possible, i.e. the pin gap at the center conductor was kept as
small as possible. It has been shown [23] that this causes coupling effects, resulting in
an undefined reference plane, and possibly even resonance effects and repeatability

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 30 –
issues. It is therefore recommended [40] to keep the size of the pin gap not smaller
than the values given in table 8.1 for typical connector geometries. As the DUT can
have a pin recession down to zero it is advised to use test port adapters that have a
recessed pin depth not smaller than the values given in table 8.1.

8.5.2.3 The simple stability test described in 8.7 is a good method to identify any stability
problems related to the test port adapters, i.e. a loose connection between test port
cable and adapter.

8.6 Connector care


8.6.1 The appropriate handling of connectors is an important aspect of good measurement
practice. A large amount of literature on the topic is available and only the most im-
portant aspects are summarized below. Further reading is recommended, see e.g.
[41, 42]. Manufacturers of precision connectors provide additional information on good
handling practice.

8.6.2 Metrology grade connectors are precision components that require special care. Over-
torquing, rotating the body of the component during connection and rough abrasive
handling should be avoided. Cleaning, visual inspections, and pin depth measure-
ments should be carried out regularly. Dirty connectors often show degraded repeata-
bility. This especially holds for small connectors. It is not only the area of the contact
zones that is sensitive to contamination (resulting in magnitude and phase changes).
Even a contaminated thread of the connecting nut might affect the impact of the torque
force and degrade the repeatability. Protruding center pins and defects might cause
damage when mated with other components. Mechanical stress on the connection
should be avoided, see 8.5.1, because it will degrade the electrical performance of the
measurement or even lead to damage.

8.7 Initial stability test


8.7.1 To test the basic stability of the VNA measurement setup before calibration, the follow-
ing simple procedures can be performed. Connect a short to each test port and knock
on the housing of the test port adapter, perform small movements of the cable end and
reconnect the short under different connector orientations to reveal instabilities near
the measurement reference plane. An unusual spread in the raw uncorrected data in
each of the above test measurements might be an indication of one or more of the
following problems: unstable cable-connector interface, structural issues with the test
port adapter or instabilities directly at the measurement reference plane, represented
by the connector interface.

8.8 Repeatability
8.8.1 It is recommended to repeat each measurement for at least four different connector
orientations. The purpose of this practice is partly to verify the specific repeatability
of a certain connector type, but also to see if any component has a stability issue. It
is not uncommon that ill-behaving components show two or more different electrical

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 31 –
Table 8.1: Recommended minimal size of pin gaps for coaxial connections at the mea-
surement reference plane to avoid resonance effects. The values are listed in metric and
imperial units.

Connector type Minimal pin gap


/10−6 m /10−4 in

1.0 mm 5 2
1.85 mm 5 2
2.4 mm 15 6
2.92 mm 10 4
3.5 mm 15 6
Type-N (50 Ω) 12 5

states imposed by the varying mechanical stress during the measurements at different
connector orientations.

8.8.2 It is important to completely disconnect the center conductor before changing the con-
nector orientation. This prevents rotational stress on the female contact fingers and
the male pin. Rotational stress might bend the contact fingers, resulting in a change
of contact point, or damage the surface plating. The operator must exercise care. Bad
alignment of the connectors, excessive pushing or pulling during mating and discon-
necting will provide extra stress on the center conductor, impacting structures or circuits
attached to it, as e.g. soldering joints or resistive elements.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 32 –
References
[1] V. Teppati, A. Ferrero, and M. Sayed, editors. Modern RF and Microwave Measurement
Techniques. Cambridge University Press, 2013. ISBN 978-1-107-03641-3.

[2] J. Dunsmore. Handbook of Microwave Component Measurements: with Advanced VNA


Techniques. John Wiley & Sons, Ltd, 2012. ISBN 978-1-119-97955-5.

[3] M. Hiebel. Fundamentals of Vector Network Analysis. Rohde & Schwarz, 2007. ISBN
978-3-939837.

[4] Keysight Technologies. Network Analyzer Basics. Electronic document. Back to Basics
Series, available at www.keysight.com.

[5] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. Evaluation of measurement
data - Supplement 2 to the "Guide to the expression of uncertainty in measurement"
- Extension to any number of output quantities, 2011. JCGM 102:2011; available at
http://www.bipm.org/en/publications/guides/.

[6] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. Evaluation of Measurement Data -
Guide to the expression of uncertainty in measurement, 2008. JCGM 100:2008; available
at www.bipm.org/en/publications/guides/gum.html.

[7] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. International Vocabulary of Metrol-
ogy - Basic and General Concepts and Associated Terms (VIM 3rd edition), 2012. JCGM
200:2012; available at http://www.bipm.org/en/publications/guides/.

[8] R. B. Marks and D. F. Williams. A General Waveguide Circuit Theory. J. Res. Natl. Stand.
Technol., 97(5):533 – 562, 1992.

[9] M. Zeier, J. Hoffmann, P. Hürlimann, J. Rüfenacht, D. Stalder, and M. Wollensack. Estab-


lishing traceability for the measurement of scattering parameters in coaxial line systems.
Metrologia, 55:S23 – S36, 2018.

[10] J. Hoffmann, J. Ruefenacht, M. Wollensack, and M. Zeier. Comparison of 1.85mm Line


Reflect Line and Offset Short Calibration. In ARFTG Conference Digest, number 76, pages
1 – 7, 2010.

[11] N. M. Ridler and A. G. Morgan. New primary reference standard for vector network anal-
yser calibration at millimetre wavelengths in coaxial line. Meas. Science and Technol.,
19:065103, 2008.

[12] K. Wong. Characterization of calibration standards by physical measurements. In ARFTG


Conference Digest, number 39, pages 53–62, 1992.

[13] K. Wong and J. Hoffmann. Improve VNA Measurement Accuracy by Including Connector
Effects in the Models of Calibration Standards. In ARFTG Conference Digest, number 82,
pages 1–7, 2013.

[14] J. Hoffmann, M. Wollensack, J. Ruefenacht, and M. Zeier. Extended S-parameters for


imperfect test ports. Metrologia, 52:121 – 129, 2015.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 33 –
[15] Keysight Technologies. Specifying Calibration Standards and Kits for Keysight Vec-
tor Network Analyzers - Application Note. Electronic document. Available at
http://literature.cdn.keysight.com/litweb/pdf/5989-4840EN.pdf.

[16] P. I. Somlo and J. D. Hunter. Microwave Impedance Measurements - IEE electrical mea-
surement series; 2. Peter Peregrinus Ltd, London, UK, 1985.

[17] D. M. Pozar. Microwave Engineering. John Wiley & Sons, Inc., 4th edition, 2011. ISBN
978-0-470-63155-3.

[18] A. Ferrero and U. Pisani. Two-port network analyzer calibration using an unknown thru.
IEEE Microwave & Guided Wave Letters, 2(12):505 – 507, 1992.

[19] METAS VNA Tools, available at www.metas.ch/vnatools.

[20] K. Wong. Receiver Linearity Measurement Using Two CW Signals. In ARFTG Conference
Digest, number 78, pages 1–5, 2011.

[21] J. V. Butler, D. K. Rytting, M. F. Iskander, R. D. Pollard, and M. Vanden Bossche. 16-term


error model and calibration procedure for on-wafer network analysis measurements. IEEE
Trans. Microwave Theory & Tech., 39(12):2211 – 2217, 1991.

[22] A. Ferrero and F. Sanpietro. A Simplified Algorithm for Leaky Network Analyzer Calibration.
IEEE Microwave & Guided Wave Letters, 5(4):119 – 121, 1995.

[23] J. Hoffmann, P. Leuchtmann, and R. Vahldieck. Pin Gap Investigations for the 1.85 mm
Coaxial Connector. In Proceedings of the 37th European Microwave Conference, num-
ber 76, pages 388 – 391, 2007.

[24] M. Garelli and A. Ferrero. A Unified Theory for S-Parameter Uncertainty Evaluation. IEEE
Trans. Microwave Theory & Tech., 60(12):3844 – 3855, 2012.

[25] B. D. Hall. Computing uncertainty with uncertain numbers. Metrologia, 43:L56 – L61,
2006.

[26] M. Zeier, J. Hoffmann, and M. Wollensack. Metas.UncLib - a measurement uncertainty


calculator for advanced problems. Metrologia, 49:809 – 815, 2012.

[27] B. D. Hall. Object-oriented software for evaluating measurement uncertainty. Meas. Sci.
Technol., 24:055004, 2013.

[28] M. Zeier, J. Hoffmann, J. Ruefenacht, and M. Wollensack. Contemporary evaluation


of measurement uncertainties in vector network analysis. tm - Technisches Messen,
84(5):348 – 358, 2017.

[29] NIST Microwave Uncertainty Framework, available at www.nist.gov/ctl/rf-


technology/related-software.cfm.

[30] Dynamic uncertainty for S-parameters, Option 015, Keysight Technologies, available
through www.keysight.com.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 34 –
[31] J. Hoffmann, P. Leuchtmann, J. Ruefenacht, and K. Wong. S-parameters of Slotted and
Slotless Coaxial Connectors. In ARFTG Conference Digest, number 74, pages 1 – 5,
2009.

[32] B. D. Hall. VNA error models: Comments on EURAMET/cg-12/v.01, 2010. ANAMET


Report 051.

[33] F. Mubarak and G. Rietveld. Uncertainty Evaluation of Calibrated Vector Network Analyz-
ers. IEEE Trans. Microwave Theory & Tech., PP(99):1 – 13, 2017.

[34] R. Judaschke, G. Wuebbeler, and C. Elster. Second-Order Error Correction of a Calibrated


Two-Port Vector Network Analyzer. In ARFTG Conference Digest, number 73, pages 1–3,
2009.

[35] G. Wuebbeler, C. Elester, T. Reichel, and R. Judaschke. Determination of the Complex


Residual Error Parameters of a Calibrated One-Port Vector Network Analyzer. IEEE Trans.
Instr. & Meas., 58(9):3238 – 3244, 2009.

[36] G. Wuebbeler, C. Elester, T. Reichel, and R. Judaschke. Determination of Complex Resid-


ual Error Parameters of a Calibrated Vector Network Analyzer. In ARFTG Conference
Digest, number 69, 2007.

[37] J. Stenarson and K. Yhland. Residual error models for the SOLT and SOLR VNA calibra-
tion algorithms. In ARFTG Conference Digest, number 69, pages 133 – 139, 2007.

[38] J. Stenarson and K. Yhland. A new assessment method for the residual errors in SOLT
and SOLR calibrated VNAs. In ARFTG Conference Digest, number 69, pages 140 – 145,
2007.

[39] M. Zeier, J. Hoffmann, P. Huerlimann, J. Ruefenacht, M. Wollensack, R. Judaschke, and


K. Kuhlmann. Stability tests of electronic calibration units. In CPEM Conference Digest,
pages 16 – 17, 2014.

[40] J. Hoffmann and P. Huerlimann. Key parameters of coaxial connector models - mechan-
ical design features and electrical properties. electronic document, 2015. available at
www.metas.ch/hf/docs.

[41] A. D. Skinner. ANAMET connector guide. available at http://www.npl.co.uk/anamet-


connector-guide, 2007. 3rd edition.

[42] J. Hoffmann and J. Ruefeneacht. How to make good coaxial connections. electronic
document, 2013. available at www.metas.ch/hf/docs.

[43] R. W. Beatty. A λg /4 waveguide standard of voltage standing-wave ratio. Electronics


Letters, 9(2):24 – 26, 1973.

[44] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. Evaluation of measurement data
- Supplement 1 to the "Guide to the expression of uncertainty in measurement" - Propa-
gation of distributions using a Monte Carlo method, 2008. JCGM 101:2008; available at
http://www.bipm.org/en/publications/guides/.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 35 –
[45] C. A. Hoer and G. F. Engen. Calibrating a Dual Six-Port or Four-Port for Measuring Two-
Ports with Any Connectors. In IEEE MTT-S International Microwave Symposium Digest,
pages 665 – 668, 1986.

[46] A. Ferrero and U. Pisani. QSOLT: a new fast calibration algorithm for two port S-parameter
measurements. In ARFTG Conference Digest, number 38, pages 15 – 24, 1991.

[47] O. Ostwald. T-Check accuracy test for vector network analyzers utilizing a Tee-junction.
Electronic document, June 1998. Rohde & Schwarz Application Note 1EZ43_0E, Available
at https://www.rohde-schwarz.com.

[48] G. F. Engen and C. A. Hoer. Thru-reflect-line: An improved technique for calibrating


the dual six-port automatic network analyzer. IEEE Trans. Microwave Theory & Tech.,
27(12):987 – 993, 1979.

[49] N. M. Ridler and M. J. Salter. An approach to the treatment of uncertainty in complex


S-parameter measurements. Metrologia, 39:295 – 302, 2002.

[50] B. D. Hall. Evaluating the measurement uncertainty of complex quantities: a selective


review. Metrologia, 53:S25 – S31, 2016.

[51] R. Willink and B. D. Hall. A classical method for uncertainty analysis with multidimensional
data. Metrologia, 39:361 – 369, 2002.

[52] R. Willink and B. D. Hall. An extension to GUM methodology: degrees-of-freedom calcula-


tions for correlated multidimensional estimates. eprint, arXiv:1311.0343 [physics.data-an],
2013.

[53] B. D. Hall. Expanded uncertainty regions for complex quantities. Metrologia, 50:490 –
498, 2013.

[54] B. D. Hall. Expanded uncertainty regions for complex quantities in polar coordinates.
Metrologia, 52:486 – 495, 2015.

[55] R. B. Marks. Formulations of the basic vector network analyzer error model including
switch-terms. In ARFTG Conference Digest, number 50, pages 115 – 126, 1997.

[56] D. Blackham. Application of weighted least squares to OSL vector error correction. In
ARFTG Conference Digest, number 61, pages 11–21, 2003.

[57] M. J. Salter, N. M. Ridler, and P. M. Harris. Over-determined calibration schemes for


RF network analysers employing generalised distance regression. In ARFTG Conference
Digest, number 62, pages 127 – 142, 2003.

[58] J. Hoffmann, P. Leuchtmann, J. Ruefenacht, and R. Vahldieck. A stable bayesian vector


network analyzer calibration algorithm. IEEE Trans. Microwave Theory & Tech., 57(4):869
– 880, 2009.

[59] M. Wollensack, J. Hoffmann, J. Ruefenacht, and M. Zeier. VNA Tools II: S-parameter
uncertainty calculation. In ARFTG Conference Digest, number 79, pages 1 – 5, 2012.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 36 –
[60] K. V. Mardia, J. T. Kent, and J. M. Bibby. Multivariate Analysis. Academic Press, 2003.

[61] F. Mubarak, M. Zeier, J. Hoffmann, N. M. Ridler, M. J. Salter, and K. Kuhlmann. Verification


concepts in S-parameter measurements. In CPEM Conference Digest, 2016.

[62] A. Lewandowski. Multi-frequency approach to vector-network-analyzer scattering-


parameter measurements. PhD thesis, Warsaw University of Technology, 2010.

[63] F. Mubarak, G. Rietveld, D. Hoogenboom, and M. Spirito. Characterizing Cable Flexure


Effects in S-parameter Measurements. In ARFTG Conference Digest, number 82, 2013.

[64] K. Yhland and J Stenarson. A simplified treatment of uncertainties in complex quantities.


In CPEM Conference Digest, pages 652 – 653, 2004.

[65] IEEE. IEEE 287-2007 Standard for Precision Coaxial Connectors (DC to 110 GHz), 2007.

[66] T. Reichel, R. H. Judaschke, and F. Rausche. Effect of Uncorrected Test Port Match of a
Vector Network Analyzer on the Transmission Coefficient Gained after SOLT Calibration.
accepted by CPEM conference, 2018.

[67] N. M. Ridler and C. Graham. Some typical values for the residual error terms of a calibrated
vector automatic network analyser (ANA). In Proc BEMC Conference, pages 45/1 – 45/4,
1999. Brighton, UK.

[68] J. Hoffmann. Errors in the Ripple Technique due to Pin Gap. electronic document, 2016.
available at www.metas.ch/hf/docs.

[69] B. D. Hall. On the expression of measurement uncertainty for complex quantities with
unknown phase. Metrologia, 48:324 – 332, 2011.

[70] T. Reichel. Derivation of the Uncertainty assigned to the Residual Reflection Track-
ing Parameter with SOL(T) calibrations. electronic document, 2018. available at
www.metas.ch/hf/docs.

[71] A. R. Kerr. Mismatch caused by waveguide tolerances, corner radii, and flange
misalignment. Technical report, National Radio Astronomy Observatory, Char-
lottesville, VA, USA, 2009. Electronics Division Technical Note No 215, available at
http://www.gb.nrao.edu/electronics/edtn/edtn215.pdf.

[72] D. J. Bannister, E. J. Griffin, and T. E. Hodgetts. On the dimensional tolerances of rect-


angular waveguide for reflectometry at millimetric wavelengths. Technical report, National
Physical Laboratory, Teddington, UK, 1989. NPL Report DES 95.

[73] N. M. Ridler and M. J. Salter. Cross-connected Waveguide Lines as Standards for


Millimeter- and Submillimeter-wave Vector Network Analyzers. In ARFTG Conference
Digest, number 81, 2013.

[74] T. Schrader, K. Kuhlmann, R. Dickhoff, J. Dittmer, and M Hiebel. Verification of scattering


parameter measurements in waveguides up to 325 GHz including highly-reflective devices.
Adv. Radio Sci., 9:9 – 17, 2011.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 37 –
[75] N. M. Ridler. Choosing line lengths for calibrating waveguide vector network analysers at
millimetre and sub-millimetre wavelengths. Technical report, National Physical Laboratory,
Teddington, UK, 2009. NPL Report TQE 5.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 38 –
A Glossary

ADC Analog to digital converter.


Air-dielectric line Air-dielectric coaxial transmission line.
Beaded air-dielectric line An air-dielectric line with a fixed center conductor that is
kept in place by one or more dielectric beads.
Beadless air-dielectric An air-dielectric line consisting of two separate parts, the
line center conductor and the outer conductor. The center
conductor is kept in place by the counterparts when the
air-dielectric line is mounted.
Beatty Line A mismatched air-dielectric line with a central section de-
viant from 50 Ω (e.g. 25 Ω) and 50 Ω sections at both
ends [43].
Cartesian coordinates Also called rectangular coordinates. The two-
dimensional Cartesian coordinate system consists of a
horizontal and a vertical axis, conventionally denoted as
x-axis and y-axis.
Directivity VNA error coefficient.
DUT Device under test.
ECU Electronic calibration unit.
Fixed load Alternative expression for matched load. Usually used to
differentiate from a sliding load.
Flush short A short without a section of transmission line and with a
reflection coefficient close to -1.
Flush thru Direct connection of two test ports, corresponding to the
flush mating of two reference planes, with nominal reflec-
tion coefficients of 0 and transmission coefficient of 1.
IFBW Intermediate frequency bandwidth. A VNA setting that
affects the noise floor of the VNA.
ISO-GUM Guide to the Expression of Uncertainty in Measurement,
consisting of a main document [6] and supplementary
documents, e.g. [44, 5]. Authoritative guidance for the
evaluation of measurement uncertainties.
Isolation VNA error coefficient.
Load match VNA error coefficient.
LRL calibration VNA two port calibration (Line-Reflect-Line) based on
measurements of highly reflective standards (open or
short) and air-dielectric lines [45].

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 39 –
Matched load Impedance matched one-port device with a reflection co-
efficient close to 0.
Mismatch Impedance mismatched one-port device. Typical com-
mercially available mismatch standards have reflection
coefficients between 0.1 and 0.5.
Offset open An open with a section of transmission line with finite
electrical length in front of it. Often just referred to as
open.
Offset short A short with a section of transmission line with finite elec-
trical length in front of the shorting plane. Often just re-
ferred to as short.
Open One port device with a reflection coefficient close to 1 at
DC.
OSM calibration See SOL calibration.
PDF Probability density function. Function that describes the
relative likelihood of a random variable to take on a given
value.
Primary experiment The process of the SI traceable characterization of pri-
mary measurement standards.
Primary standard Measurement standard that can be characterized
through a primary experiment. Typical primary standards
are air-dielectric lines, offset shorts, flush shorts and off-
set opens.
Reflection tracking VNA error coefficient.
Rigorous Method Uncertainty evaluation of VNA measurements based on
rigorous modeling of the entire VNA measurement pro-
cess.
Ripple Method Uncertainty evaluation of VNA measurements based on
the determination of residual VNA error coefficients. The
residual error coefficients are determined through the
analysis of measured ripple patterns.
S-parameters Scattering parameters.
Short One port device with a reflection coefficient close to -1 at
DC. Two types of shorts are typically distinguished: offset
short and flush short.
Signal flow graph Graphical representation of a linear network, consisting
of branches and nodes.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 40 –
Sliding load One-port device consisting of an air-dielectric line termi-
nated with a matched load. The load is an integral part
of the air-dielectric line and can be moved in the longitu-
dinal direction within the air-dielectric line, thus changing
the phase relation of the reflected signal.
SOL calibration VNA one port calibration based on short, open and
(matched) load standards. Sometimes also called OSM
or OSL calibration.
SOLR calibration Short-Open-Load-Reciprocal VNA two port calibration
[18]. Also known as Unknown Thru calibration.
SOLT calibration VNA two port calibration based on short, open,
(matched) load standards and the measurement of a
flush thru connection. Sometimes also called OSMT cal-
ibration.
Source match VNA error coefficient.
QSOLT calibration Quick SOLT. VNA two port calibration method based on
the transfer of the calibration from one port to the other
[46].
SSS calibration VNA one port calibration based on offset shorts.
SSST calibration VNA two port calibration based on offset shorts and a
flush thru measurement.
T-checker Two-port verification device based on a T-junction, one
port of which is terminated by a load impedance [47].
TOSM calibration See SOLT calibration.
Transmission tracking VNA error coefficient.
TRL calibration VNA two port calibration based on measurements of
flush thru connection, reflects (open or short) and air-
dielectric lines [48].
Unknown Thru calibration see SOLR calibration.
VNA Vector network analyzer.
VNA calibration Determination of VNA error coefficients and VNA switch
terms by the measurement of a set of VNA calibration
standards.
VNA calibration standards Characterized measurement standards, which are used
during VNA calibration to determine error coefficients
and switch terms.
VNA error model Model of the systematic errors in a VNA measurement.
It can be represented graphically as a signal flow graph
of a linear network.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 41 –
VNA error coefficients Coefficients in the VNA error model, which are deter-
mined during VNA calibration.
VNA isolation VNA error coefficient describing cross-talk effects.
VNA switch term VNA error coefficient describing the effect when the
source is switched to the other port.
VNA measurement model VNA error model extended with influences that are af-
fecting the measurement. Represents the basis for the
evaluation of the measurement uncertainty.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 42 –
B Notation
The quantities in VNA measurements are typically complex-valued. They are written
in bold face, such as S11 . Real and imaginary component are represented by Re[S11 ]
and Im[S11 ], respectively. Magnitude and phase are represented by |S11 | and arg(S11 ),
respectively. Vectors and matrices are written in bold face whereas scalar values are
printed in plain face.
When properly dealing with measurement quantities and uncertainties one has to
distinguish between the quantity itself and an estimate of the quantity. Quantities are
denoted by plain characters, e.g. S11 , whereas estimates are accented with a hat, e.g.
Ŝ11 .
Quantity estimates (and measured values) are never exactly equal to the quantities
themselves; there is always some measurement error involved. For that reason, an un-
certainty is associated with each estimate of a quantity. In the scalar case uncertainties
are denoted by a plain faced u(·), e.g. u(Re[Ŝ11 ]) and in the complex-valued or vector
case by a bold faced u(·), e.g. u(Ŝ11 ). In the latter case the uncertainty is represented
by a matrix, as shown in appendix C.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 43 –
C S-parameter measurement uncertainties
A basic framework for the evaluation of measurement uncertainty of S-parameters has
been described in [49]. This was already largely in agreement with the treatment of
multidimensional (or multivariate) quantities promoted in Supplement 2 [5] of the ISO-
GUM [6], which was published more than a decade later.
The purpose of this section is to provide a brief introduction to S-parameter mea-
surement uncertainties and point out some aspects that are relevant in practical mea-
surements. This is by no means comprehensive and further information should be
obtained from literature. Recommended reading is [5] for the general treatment of
multivariate measurement uncertainties and [50], which is more specifically related to
complex-valued quantities.

C.1 S-parameter data format


S-parameters are complex-valued quantities and can be represented either by polar
coordinates (magnitude and phase) or Cartesian coordinates (real and imaginary com-
ponents). Polar coordinates are directly related to some of the physical phenomena
occurring during a measurement and are therefore sometimes preferred for the inter-
pretation and understanding of results. It has been demonstrated, however, that polar
coordinates might cause problems when used in statistical analysis or for uncertainty
evaluation [49]. It is therefore recommended to perform any calculations in Cartesian
coordinates. For the interpretation the data can be transformed to polar coordinates.
Especially for the reflection coefficient of highly reflective devices the visualization in po-
lar coordinates is usually more meaningful, whereas well matched devices are prefer-
ably studied in Cartesian coordinates.
The following sections will therefore use Cartesian coordinates only, though the for-
malism is in principle also applicable to polar coordinates if the necessary precautions
are taken.

C.2 Representation of measurement uncertainties


A quantity of interest (the measurand) cannot be determined exactly because of mea-
surement error. Therefore, the measured value can only be regarded as an estimate of
the measurand, the unknown error being covered with the stated uncertainty.
Uncertainty associated with the measurement of a quantity of interest is character-
ized by a probability density function (PDF). For an S-parameter, or any other complex-
valued quantity, a two-dimensional PDF is used. The type of distribution and a scale
factor for the extent of the distribution will be chosen to represent the uncertainty of
measurement. For a real-valued quantity, the so-called “standard uncertainty” is the
standard deviation (square root of the variance) of the PDF. In the case of complex
quantities (and indeed for other multidimensional quantities), the variances and covari-
ances of the chosen PDF play a similar role.
This is formally shown for a single complex-valued quantity x (e.g. one of the S-
parameters). The variable x̂ denotes the estimate of that quantity. The measurement
uncertainty associated with x̂ depends on the real and imaginary components and is

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 44 –
conveniently represented by a symmetric uncertainty matrix, which is sometimes also
called covariance matrix: !
u11 u12
u(x̂) = , (C.1)
u21 u22
where

u11 = (u(Re [x̂]))2


u22 = (u(Im [x̂]))2 (C.2)
u12 = u21 = r (Re [x̂] , Im [x̂]) u(Re [x̂]) u(Im [x̂]) ,

with u(Re [x̂]) and u(Im[x̂]) the standard uncertainties associated with the real and
imaginary components and r(Re[x̂], Im[x̂]) the correlation coefficient between the real
and imaginary component estimates.

C.2.1 Correlation
Correlation in measured values is caused by common influences. A few examples: the
use of the same reference standard for the characterization of calculable calibration
standards will cause correlations between the mechanical measurements of calculable
standards. If these standards are used to characterize transfer standards, i.e. an SOL
kit, short, open and matched load will be correlated. The measurements of a short stan-
dard during a VNA one-port calibration will influence the determination of the reflection
tracking and the source match resulting in correlation between them. The multiplication
of measured values for two complex-valued quantities (i.e. two reflection coefficients)
in Cartesian coordinates will lead to correlation between the real and imaginary com-
ponents of the product, even if the initial measurements were uncorrelated (i.e. the
quantity estimates are not correlated).
Correlation in VNA measurements is unavoidable, and it can either increase or
decrease the measurement uncertainty. The best way to take correlation effects into
account is through refined modeling of the measurement process. Common influence
quantities are part of the measurement model. Rigorous uncertainty propagation based
on such a measurement model, see 7.1, will take correlation properly into account. For
random effects it is possible to evaluate correlation in a statistical way by using the
formalism in C.5.

C.3 Evaluation of measurement uncertainty


Uncertainty analysis requires a mathematical model, the so-called measurement model,
which relates the quantity of interest y, e.g. the S-parameter of a DUT, to input quantities
x1 , x2 , ... that influence the outcome of the measurements, e.g. the characterization of
VNA calibration standards or VNA device parameters as noise, non-linearity and drift:

y = f (x1 , x2 , ...) . (C.3)

The terms that appear in a measurement model (C.3) refer to actual quantities. But
as pointed out in C.2 there are only estimates available. An estimate of the quantity

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 45 –
of interest is obtained by evaluating the measurement model with the estimates of the
input quantities.
The measurement uncertainty associated with the estimate (measured value) of the
quantity of interest y is evaluated in three steps.
1. Determination of the estimates x̂1 , x̂2 , ... and associated uncertainties u(x̂1 ),
u(x̂2 ), ... of the input quantities x1 , x2 , ...: characterization of input quantities
by statistical analysis, systematic studies, previous evaluations, device specifica-
tions, etc. The ISO-GUM [6] distinguishes between Type A (statistical analysis),
see remarks in C.5, and Type B (by any other means) evaluations.
2. Calculation of uncertainty of the quantity of interest: propagation of the estimates
x̂1 , x̂2 , ... and uncertainties u(x̂1 ), u(x̂2 ), ... of the input quantities x1 , x2 , ...
through the measurement model (C.3) to obtain an estimate ŷ of the quantity of
interest y, with an associated uncertainty u(ŷ) . See remarks in C.4.
3. Scaling of u(ŷ) by a coverage factor k to obtain an expanded uncertainty with a
desired coverage probability. This is typically 95%. See remarks in C.6.

C.4 Uncertainty propagation


To propagate the uncertainties associated with estimates of the input quantities x1 , x2 ,
... to the quantity of interest y through the measurement model (C.3) two established
techniques are available. Both techniques have the capability to take into account cor-
relations between the estimates of the input quantities.
1. Linear uncertainty propagation: this method is described in GUM Supplement 2
[5] and is a generalization of the scalar case described in [6]. Variances and
covariances of the input quantities are propagated through a linearized measure-
ment model. The uncertainty in the measured value of the quantity of interest
is assumed to be characterized by a multivariate Gaussian PDF. The variance-
covariance of this PDF is found by propagating variance-covariance information
associated with the input quantity estimates through the measurement model.
The result can be obtained through an analytical calculation.
2. Monte Carlo method: this method is described in GUM Supplement 2 [5] and
generalizes the scalar method in GUM Supplement 1 [44]. Here the distributions
are propagated through the measurement model by making repeated draws from
the PDFs of the input quantities. The result is a PDF of the quantity of interest.
Numerical methods with the generation of pseudo-random numbers are usually
needed to obtain the result.
Linear uncertainty propagation makes some simplifying assumptions to allow for an
analytical calculation. In the large majority of the cases the results are nevertheless
accurate enough. For complex-valued quantities there are some cases known where
linear uncertainty propagation can fail, e.g. when transforming values near zero be-
tween Cartesian and polar coordinates. The computational effort is, however, generally
significantly larger for the Monte Carlo method compared to linear uncertainty propa-
gation. This is a criterion for VNA measurements, which often consist of sweeps over

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 46 –
several hundred frequency points. In linear uncertainty propagation, sensitivity coeffi-
cients (derivatives) are calculated. With this information it is easy to provide an uncer-
tainty budget. An uncertainty budget is valuable to understand the significance of the
different influence factors affecting the measurement. The compilation of an uncertainty
budget with the Monte Carlo method increases the already large computational effort
significantly. For these reasons it is often preferable to use linear uncertainty propaga-
tion for VNA measurements. The Monte Carlo method can still be used for comparison
purposes.

C.5 Sample statistics and Type A uncertainty


C.5.1 Sample statistics of scalar quantities
When a scalar quantity x is estimated from a sample of n observations xm (1), xm (2),
..., xm (n) the sample mean is
n
1X
m(x) = xm (i) . (C.4)
n
i=1

The sample variance of the n observations is


n
1 X
v(x) = (xm (i) − m(x))2 . (C.5)
n−1
i=1

The sample variance of the mean of the n observations is


v(x)
v(m(x)) = . (C.6)
n
Closely related to the variance is the standard deviation of the n observations
p
s(x) = v(x) (C.7)

and the standard deviation of the mean


p s(x)
s(m(x)) = v(m(x)) = √ . (C.8)
n

C.5.2 Sample statistics of complex-valued quantities


The scalar formalism can be generalized for a complex-valued quantity, as e.g. an S-
parameter. If the complex-valued quantity x = Re[x] + j Im[x], with the imaginary unit j,
is estimated from a sample of n observations xm (1), xm (2), ..., xm (n) the sample mean
is
m(x) = m(Re [x]) + jm(Im [x]) (C.9)
with the scalar mean m(·) according to equation (C.4). The covariance matrix of the n
observations is !
v11 v12
v(x) = , (C.10)
v21 v22

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 47 –
where

v11 =v(Re [x])


v22 =v(Im [x])
n
1 X
v12 = v21 = (Re [xm (i)] − m(Re [x])) (Im [xm (i)] − m(Im [x])) ,
n−1
i=1

with the scalar variance v(·) and the scalar mean m(·) according to equations (C.5) and
(C.4), respectively. The covariance matrix of the mean is accordingly
1
v(m(x)) = v(x) . (C.11)
n
The formalism for complex-valued quantities is defined in direct analogy to the scalar
case with the notable addition of the correlation information, which is contained in v21 in
equation (C.10). The same principle formalism holds for vectors of length N , resulting
in a mean vector of length N for scalar quantities (2N for complex-valued quantities)
and covariance matrix of size N × N for scalar quantities (2N × 2N for complex-valued
quantities).

C.5.3 Type A uncertainty


The statistical analysis in C.5.1 and C.5.2 can be used to assig measurement uncer-
tainties based on repeated measurements of a quantity. In the ISO-GUM [6] this is
referred to as Type A evaluation of a measurement uncertainty. The discussion below
is limited to uncertainties used in linear uncertainty propagation. For the use with the
Monte Carlo method information can be found in [44, 5].

C.5.3.1 Large sample size


If the length n of the sample of observations is large the results in C.5.1 and C.5.2 can
be directly used. For a scalar quantity x the mean according to equation (C.4) is the
best estimate x̂ with an uncertainty u(x̂)

u(x̂) = s(m(x)) , (C.12)

with the standard deviation of the mean s(m(x)) according to equation (C.8).
For a complex-valued quantity x the mean according to equation (C.9) is the best
estimate x̂ with an uncertainty matrix u(x̂)

u(x̂) = v(m(x)) (C.13)

with the covariance matrix of the mean v(m(x)) according to equation (C.11).

C.5.3.2 Small sample size


If the length n of the sample of observations is small, applying equations (C.12) and
(C.13) would underestimate the measurement uncertainty. The ISO-GUM documents
are currently not consistent in the way they deal with such a situation.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 48 –
The ISO-GUM [6] recommends to use uncertainties according to equations (C.12)
and (C.13) together with degrees of freedom, which are defined as n − 1, with n the
size of the measurement sample. This information is later used to derive expanded
combined uncertainties, see C.6.
GUM Supplement 2 [5] recommends to increase the uncertainty, which is derived
from the statistical analysis, by a factor [(n − 1)/(n − N − 2)]1/2 , with n the size of
the measurement sample and N the dimension of the quantity. For scalar quantities
(N = 1) this leads to r
n−1
u(x̂) = s(m(x)) , (C.14)
n−3
and for complex-valued quantities (N = 2) to
n−1
u(x̂) = v(m(x)) . (C.15)
n−4
These uncertainties are directly used in the subsequent measurement uncertainty prop-
agation and there is no need to use degrees of freedom to derive expanded combined
uncertainties, see C.6. However, to apply equations (C.14) and (C.15) at least four
(scalar quantity) or five (complex-valued quantity) measurement repetitions are neces-
sary. If say one cares about correlation between all four S-parameter of a two-port
measurement (N = 8), then at least 11 measurement repetitions would be needed.

C.5.3.3 Other considerations


Assigning measurement uncertainties based on a Type A analysis as described in
C.5.3.1 or C.5.3.2 assumes that the quantity has a time invariant and unique true value.
If the value of the quantity is drifting over time or if the value changes discretely, e.g. a
coaxial connection showing two or more states under different connector orientations,
blindly applying the Type A formalism would lead to an underestimation of the measure-
ment uncertainty.
A large number of measurement repetitions is not always feasible. The effort needs
to be considered with respect to the contribution of the Type A uncertainty to the overall
measurement uncertainty. If the contribution is small a reduced number of repetitions
is sufficient. If effects described in the previous paragraph can’t be excluded and a
large number of measurement repetitions to study the effects is not feasible, it is rec-
ommended to take a more conservative approach and e.g. take the half-width between
maximum and minimum value of the sample of observations as the measurement un-
certainty. Personal judgement is advised in such cases.

C.6 Expanded uncertainty


In the final step of any evaluation of measurement uncertainty the standard uncertainty
of the quantity of interest is often multiplied by a coverage factor k to obtain an ex-
panded uncertainty, which has a desired coverage probability, typically 95%. The exact
meaning of this coverage depends on the interpretation of probability. Up to now the
ISO-GUM documents are not consistent in that respect. This is related to the inconsis-
tency mentioned previously in C.5.3.2. The discussion below is limited to uncertainties

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 49 –
obtained with linear uncertainty propagation, i.e. it is assumed that the PDF associated
with the measured value of the quantity of interest has a Gaussian shape. Expanded
uncertainties based on the Monte Carlo method are described in [44, 5].

C.6.1 Method according to GUM Supplement 2 [5]: if the Type A uncertainties are calculated
with the equations (C.14) and (C.15), the coverage factors for the combined uncertainty
are k = 1.96 for 95% coverage in the scalar case and k = 2.45 for 95% coverage in
the two-dimensional case. The latter can be applied to a complex-valued quantity and
corresponds to an elliptical region in the complex plane.

C.6.2 If the Type A uncertainties are calculated according to equations (C.12) and (C.13) and
propagated to the combined uncertainty, it is necessary to calculate effective degrees
of freedom by applying the Welch-Satterthwaite formula. The coverage factor is then
dependent on the effective degrees of freedom. This is described in the ISO-GUM
[6] for the case of uncorrelated scalar quantities. A generalization of the procedure to
multivariate and complex-valued quantities is described with an example in [50]. Further
details are given in [51, 52].

C.6.3 The discrepancies between the two methods described in C.6.1 and C.6.2 become
apparent when Type A uncertainties based on a small sample sizes (as discussed in
C.5.3.2) make a significant contribution to the overall uncertainty. Otherwise the two
approaches converge to the same coverage factors. The method in C.6.1 is easier to
apply, but there are indications [52] that it provides overly conservative coverage regions
compared to the method in C.6.2.

C.6.4 The expanded uncertainty of a complex-valued quantity corresponds to a region in the


complex plane. It is common to use an elliptical region, because this corresponds to
the contour of a two-dimensional Gaussian PDF. Alternative shapes are discussed in
[53, 54].

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 50 –
D VNA calibration
In this section, further details are given for commonly used VNA calibration schemes.
The discussion is limited to methods that are relevant (not exclusively) for coaxial mea-
surements. It is not the intent to give a complete overview on all existing VNA calibration
algorithms.

D.1 One-port calibration techniques


For one-port DUT measurements, the one-port error model, illustrated as a signal flow
graph in figure D.1, is used. It contains the three error coefficients in table D.1 relating
the reflection coefficient indicated by the VNA, Sm11 , to the true reflection coefficient S11
of the device being measured. Sm 11 is sometimes also referred to as a raw S-parameter.
Evaluation of the signal flow graph leads to
E01 S11
Sm
11 = E00 + . (D.1)
1 − E11 S11
Three known reflection standards are necessary to determine estimates of the error co-
efficients. Measuring the three standards successively gives three independent equa-
tions, according to equation (D.1), with the three error coefficients as the unknown
variables. The three equations can be linearized through parameter substitution and
solved by classical linear algebra techniques. Solving equation (D.1) for S11 leads to
Sm
11 − E00
S11 = , (D.2)
E11 (Sm
11 − E00 ) + E01

which is used to perform the error correction on the DUT measurement. The previ-
ously determined estimates of the error coefficients are substituted into equation (D.2)
together with Sm11 for the DUT to obtain an estimate of the S-parameter, Ŝ11 , for the DUT.
Ŝ11 is sometimes also referred to as the error corrected S-parameter.
The most common one-port calibration method for coaxial measurements is based
on a short, an open and a matched load as calibration standards. Common acronyms
are SOL (short - open - load) or OSM (open - short - match).
An alternative to SOL is to use a set of shorts with different offset lengths. However,
this technique is band-limited and more than three offset shorts are necessary to cover
larger frequency ranges. The acronym for this technique is SSS. It is based on calcu-
lable standards and can therefore be used for primary experiments, see the note on

Table D.1: List of error coefficients of one-port error model.

Symbol Error coefficient

E00 Directivity
E01 Reflection tracking
E11 Source match

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 51 –
1
a1

E00 E11
S11

E01
b1

Figure D.1: Signal flow graph representing a one port VNA error model with the error
coefficients listed in table D.1, which relate indicated reflection coefficient Sm
11 = b1 /a1
with true reflection coefficient S11 .

SSST in D.2.2. An advantage of SSS is that shorts are typically more robust and stable
than other types of standards. A disadvantage is the bandwidth limitation. It is possible
to overcome the bandwidth limitation at low frequencies by using additional open and
matched load standards.

D.2 Two-port calibration techniques


D.2.1 Introduction
For two-port DUT measurements, two different error models are commonly used, the
seven-term error model and the ten-term error model. These two models are tradi-
tionally related to two different VNA architectures, the four-receiver VNA or dual re-
flectometer VNA, where each reflectometer has its own reference channel, and the
three-receiver VNA with a single reference channel. These architectures are shown in
figure D.2. For further details see e.g. [1, 3]. The two error models and the different
two port calibration algorithms are related to each other. In [55, 1] it is discussed under
which conditions a transformation from one model to the other can be performed.

D.2.2 Ten-term error model


The ten-term error model has been developed for three-receiver VNAs but is also ap-
plicable to four-receiver VNAs. This model assumes that each port has two different
states:

• In the first state, the test port is connected to the source and two receivers mea-
sure the incoming and outgoing waves. This corresponds to the configuration of
one-port measurements and is characterized by the three error coefficients, di-
rectivity, source match and reflection tracking, which can be determined with a set
of three one-port standards, as discussed in D.1.

• In the second state, the port is connected to the switch termination, and incoming
waves are measured by only one receiver. This state is modeled with two error

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 52 –
a0

a1 a2

b1 b2 b1 b2

Port 1 Port 2 Port 1 Port 2


Figure D.2: Four-receiver (left) vs. three-receiver (right) VNA architecture. ai and bi de-
note the wave quantities measured by the reference and test receivers, respectively.

coefficients, the load match and the transmission tracking, which can be deter-
mined by the measurement of a two-port standard. With a four-receiver VNA, the
corresponding reference channel is not used.

Figure D.2 (right side) shows port 1 in the first state and port 2 in the second state.
Changing the switch will put port 2 into the first state and port 1 into the second state.
This results in two error models. The forward model (figure D.3), relates indicated and
true forward S-parameters, Sm m
11 , S21 and S11 , S21 , respectively. The reverse model
(figure D.4), relates indicated and true reverse S-parameters, Sm m
22 , S12 and S22 , S12 ,
respectively. The error coefficients are listed in table D.2. The ten-term error model
is also referred to as twelve-term error model, if isolation terms are also taken into
account.

Table D.2: List of error coefficients of forward and reverse ten-term error model.

Symbol Error coefficient


forward reverse

EF00 ER
33 Directivity
EF01 ER
32 Reflection tracking
EF11 ER
22 Source match
EF22 ER
11 Load match
EF32 ER
01 Transmission tracking
EF30 ER
03 Isolation

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 53 –
EF30

1 S21 EF32
a0 b2

EF00 EF11 S11 S22 EF22

EF01 S12
b1

Figure D.3: Signal flow graph representing the ten-term forward VNA error model (in-
cluding the isolation term) for two port measurements with the error coefficients listed in
table D.2. The model relates indicated forward S-parameters, Sm m
11 = b1 /a0 and S21 = b2 /a0 ,
with true S-parameters, S11 and S21 .

S21 ER
32
b2

ER
11 S11 S22 ER
22 ER
33

ER
01 S12 1
b1 a0
ER
03

Figure D.4: Signal flow graph representing the ten-term reverse VNA error model (includ-
ing the isolation term) for two-port measurements with the error coefficients listed in
table D.2. The model relates indicated reverse S-parameters, Sm m
22 = b2 /a0 and S12 = b1 /a0 ,
with true S-parameters, S22 and S12 .

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 54 –
The determination of the error coefficients of the ten-term error model is based on
an SOL calibration at each port with an additional measurement of a flush thru. The
flush thru is realized by directly connecting the two test ports. It is defined as a perfect
connection of both reference planes with zero length in-between. Common acronyms
for this calibration method are SOLT and TOSM.
NOTE — In principle, a characterized transmission line or adapter can be used instead of
the flush thru standard in the SOLT calibration. However, this generally degrades the accuracy.
As described for the one-port case, the two-port offset short calibration is an al-
ternative as well. The acronym is SSST. SSST is based on calculable standards and
can therefore be applied for primary experiments. It is an alternative to the LRL (line -
reflect - line) calibration at higher frequencies when the handling of air-dielectric lines
becomes more difficult, see D.2.3.

D.2.3 Seven-term error model


The signal flow graph of the seven-term error model is shown in figure D.5. It includes
seven (nine, if the isolation terms are taken into account) unknown error coefficients
and two switch terms listed in table D.3. This error model is sometimes called the eight-
term error model, ignoring the fact that one of the tracking terms is set to 1 due to
normalization.
Several calibration techniques based on the seven-term error model have been de-
veloped, using any sufficient combination of one-port and two-port standards. The most
popular calibration methods are short-open-load-reciprocal (SOLR), also known as Un-
known Thru [18], thru-reflect-line (TRL) [48], line-reflect-line (LRL) [45] and QSOLT [46].
For the determination of the switch terms, and thus for the use of VNA calibrations
that are based on the seven-term error model, it is in principle necessary that the VNA
has a four-receiver architecture. There is however a way to overcome this limitation.
With the help of an additional SOLT calibration it is possible to determine the switch
terms computationally. The theory is based on [55]. In some VNAs proprietary imple-
mentations of this method are available in the firmware.

Table D.3: List of error coefficients of seven-term error model.

Symbol Error coefficient


port 1 port 2

E00 E33 Directivity


E01 E32 E23 Reflection tracking
E11 E22 Source match
E30 E03 Isolation
W00 W33 Switch term

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 55 –
E30

a1 1 S21 E32 b2

W00 E00 E11 S11 S22 E22 E33 W33


E01 S12 E23
b1 E03 a2

Figure D.5: Signal flow graph representing the seven-term VNA error model (including
isolation terms) for two port measurements with the error coefficients and the switch
terms listed in table D.3. The model relates indicated S-parameters, Sm m
11 = b1 /a1 , S21 =
m m
b2 /a1 , S22 = b2 /a2 and S12 = b1 /a2 , with true S-parameters, S11 , S21 , S22 and S12 .

D.3 Over-determined calibration


More recently, optimization calibration algorithms have been proposed, where more
calibration standards than needed are measured. In any of the previously mentioned
calibration algorithms it is possible to measure more than the minimum number of cal-
ibration standards required to determine the error coefficients. This results in an over-
determined system of equations which can be solved by an optimization algorithm, e.g.
a least squares routine. Different implementations are known [56, 57, 58, 59] and some
commercial VNA firmware supports over-determined calibration as well, e.g. in combi-
nation with electronic calibration units or with offset short calibration kits.
Using more than the necessary number of standards potentially increases the ac-
curacy of the calibration and reduces the measurement uncertainty. It provides an
excellent diagnostic tool to detect problems (drift or change) of single calibration stan-
dards. For this purpose it is necessary to calculate the residuals of the optimization,
i.e. the difference between the reference data and the corrected measurements of the
calibration standards. If one of the standards shows a significant larger difference than
the others, further investigations are advised.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 56 –
E VNA verification
This section provides additional information related to VNA verification. A discussion
of different verification standards and their relation to VNA error coefficients is given in
E.1. Quantitative pass/fail criteria for verification measurements are discussed in E.2.

E.1 Verification standards


Depending on the type and characteristics of the device under test, different verification
artifacts can be used to verify the calibration of a VNA. The overall Smith chart region,
for both one- and two-port measurements, cannot be completely covered by one single
verification standard over the entire frequency range. For a full coverage of the Smith
chart several different verification standards are needed, as described below. Proximity
to the DUT response increases the significance of a single verification measurement.

E.1.1 One-port verification standards


To achieve confidence in reflection measurements, calibrated matched, mismatched,
and highly reflective one-port artifacts (with both male and female interfaces) can be
measured. A sensitivity analysis can be based on the residual measurement model,
which is introduced in F.5. Ignoring all influence terms except the residual error coeffi-
cients and linearizing with respect to them [32] leads for the one-port case to

Sc11 = δE00 + (1 + δE01 ) S11 + δE11 S211 . (E.1)

This equation relates the error corrected reflection coefficient Sc11 to the true reflection
coefficient S11 of the DUT via the residual one-port error coefficients, δE00 (directivity),
δE11 (source match) and δE01 (reflection tracking). Since the sensitivity of Sc11 to the
individual residual error coefficients depends on the reflectivity of the DUT, S11 , equa-
tion (E.1) might serve as a guideline to choose verification standards appropriately to
identify potential sources of erroneous calibration.

E.1.1.1 Matched load standard


With a matched load the central part of the Smith chart is being tested (|S11 | ' 0).
According to equation (E.1) the measurement of a matched load is directly and exclu-
sively sensitive to the residual directivity. A load measurement helps to identify prob-
lems related to misuse or defective behavior of a fixed load (or sliding load) during VNA
calibration.

E.1.1.2 Highly reflective standard


Following equation (E.1), the measurement of highly reflective standards (|S11 | ' 1) are
sensitive to all residual error coefficients and help to identify problems related to misuse
or defects of highly reflective standards used during VNA calibration.
A flush short is particularly useful to test how well the reference plane has been
defined during the VNA calibration, as pointed out in 4.3.2.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 57 –
E.1.1.3 Mismatch standard
Mismatch standards with e.g. 0.2 < |S11 | < 0.5 are suitable to cover the intermediate
regions of the Smith chart. Problems related to non-linearity, e.g. a receiver operated in
compression mode, can be detected with such a verification measurement.

E.1.2 Two-port verification standards


In analogy to the one-port case equations can be derived to express the sensitivities of
Sc11 and Sc21 with respect to the 2-port residual error coefficients [32]. This is based on
a linearization of the two-port residual measurement model and leads to

Sc11 = δEF00 + 1 + δEF01 S11 + δEF11 S211 + δEF22 S21 S12



(E.2)
Sc21 = δEF11 S11 S21 + δEF22 S22 S21 + 1 + δEF32 S21 .

(E.3)

These equations relate the error corrected S-parameters, Sc11 and Sc21 , to the true S-
parameters S11 and S21 of the DUT via the residual error coefficients, δEF00 (directivity),
δEF11 (source match), δEF01 (reflection tracking), δEF22 (load match) and δEF32 (transmis-
sion tracking). Analogous equations exist for the reverse residual error model

E.1.2.1 Beadless air-dielectric line


Ideal beadless air-dielectric lines theoretically provide the best possible impedance ref-
erence, since they can be both fabricated and mechanically characterized with high
precision. In reality, beadless air-dielectric lines are not ideal. They are not lossless,
their connectors cause reflections and the fact that the position of the center conductor
is not fixed impairs reproducibility, see [23]. These effects become more pronounced at
higher frequencies. To use beadless air-dielectric lines as accurate verification devices
they need to be characterized and the position of the center conductor needs to be
controlled [10]. Their handling, especially for the higher frequency connector systems,
demands a skilled operator. Despite these limitations, beadless air-dielectric lines are
used to estimate residual source match and residual directivity with the Ripple Method,
as described in 7.2.

E.1.2.2 Beaded air-dielectric line or adapter


Characterized beaded air-dielectric lines or adapters are a good alternative to beadless
air-dielectric lines. They offer excellent reproducibility and are easier to handle. As
weakly reflecting nearly transparent devices (|S11 |, |S22 | ' 0 and |S21 |, |S12 | ' 1) they
are sensitive to residual load match, directivity and transmission tracking, as seen from
equations (E.2) and (E.3), respectively.

E.1.2.3 Matched attenuation device


Measurements of matched attenuation devices with small to medium attenuation val-
ues, e.g. 3 dB to 20 dB, are useful to test the non-linearity of VNA receivers with two-
port measurements. For large attenuation values on the other hand, the sensitivity of
the reflection coefficients with respect to the residual errors approaches the behavior
of a one-port matched load, see E.1.1.1. Matched attenuation devices typically show
good stability because the center conductor is supported by two beads. Such a design

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 58 –
is more robust and therefore less affected by deviations in concentricity of the coaxial
structure in the mating counterpart. A matched attenuation device with a value of 30
dB or higher might therefore be used as an alternative for a matched load.
NOTE — Differences in S21 and S12 are theoretically possible for attenuation devices made
with electrically anisotropic materials. However, in practice most attenuation devices are recip-
rocal and if differences in S21 and S12 are observed they will be related to drift or to problems
with the VNA calibration.

E.1.2.4 Mismatched air-dielectric line (Beatty Line)


Transmission accuracy can be checked using a mismatched verification standard as
well. The impedance-stepped air-dielectric line invented by Beatty [43] includes a 25 Ω
line section, see figure E.1, which performs a quarter-wavelength transformation at all
odd integer multiples of the frequency f0 with f0 = c/(4L) where c denotes the ve-
locity of light and L the length of the 25 Ω section. Figure E.2 shows the simulated
S-parameter magnitudes of a typical Type-N Beatty standard. At even integer multiples
of the frequency f0 the low impedance section acts as a half-wave transformer, virtually
connecting the two ports of the VNA by a low-reflect, low-loss line. At these frequen-
cies (e.g. 2.14 GHz) the Beatty line should have low reflection, typically |S11 | < 0.01
(<-40 dB), and exhibit low transmission loss, typically |S21 | > 0.99 (>-0.1 dB). At odd
integer multiples of f0 , quarter-wavelength transformation leads to a transformation of
the 50 Ω input impedance of the VNA to 12.5 Ω, |S11 | ' 0.6 (-4.44 dB), and hence a
decreased transmission coefficient, |S21 | ' 0.8 (−1.94 dB). A Beatty line combines the
behavior of a transparent device (adapter or beaded air-dielectric line) with a one-port
mismatch standard and an attenuation device.
Beadless Beatty Lines suffer from the same issues as beadless air-dielectric lines
(E.1.2.1). For quantitative verification only characterized beaded Beatty Lines should
be used.

E.1.2.5 T-checker
An quick and easy test of VNA calibration validity utilizes a common T-junction, one
port of which is terminated by a load impedance, Z [47] as shown in figure E.3. The

50 Ω 25 Ω 50 Ω

Figure E.1: Schematic of a mismatched air-dielectric line (Beatty Line).

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 59 –
0 0

−10 −0.5

−20 −1
|S11 | / dB

|S21 | / dB
−30 −1.5

−40 −2

−50 −2.5

|S11 | |S21 |
−60 −3
0 5 10 15
Frequency /GHz

Figure E.2: Simulated S-parameter magnitudes of an ideal Type-N Beatty line with a 25 Ω
section length of 70 mm.

1 2

Figure E.3: Equivalent circuit of a T-checker.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 60 –
so-called T-check parameter cT is expressed as
|S11 S∗21 + S12 S∗22 |
cT = r   (E.4)
1 − |S11 |2 − |S12 |2 1 − |S21 |2 − |S22 |2

with ∗ denoting the conjugate transpose. The S-parameters in the equation refer to
measurements at the ports labelled 1 and 2 in figure E.3. If the T-junction is lossless,
cT is unity, independent of the load impedance Z, which can be chosen arbitrarily.
If the lossless condition is fulfilled, which might be approximately the case at low
frequencies (< 6 GHz), deviation of cT from unity indicates measurement inaccuracies
of the VNA under investigation. As a rule of thumb, deviations in the range 1 ± 0.1
can be regarded as minor. Since the T-check is sensitive to both one- and two-port
error coefficients, it enables an overall verification; however, it might not help to identify
inaccuracies of individual error coefficients.
For a truly quantitative statement based on cT = 1 the losslessness of the T-junction
needs to be demonstrated and uncertainties associated with the experimental and theo-
retical parameter cT need to be determined. Unless this is done the T-checker provides
merely a qualitative verification test.
A characterized T-checker can be used for quantitative verification as can any other
stable two-port device. Reflection and transmission coefficients of commercially avail-
able T-checker devices are typically in the order of 0.2 to 0.4 and -2 dB to -5 dB, respec-
tively.
E.1.2.6 Flush thru
After an SOLR calibration it is most useful to measure a flush thru by directly connecting
the two reference planes, provided that the DUT is insertable. Ideally, this measurement
should by definition result in S11 = S22 = 0 and S21 = S12 = 1, allowing for measure-
ment uncertainties. In order not to undermine the SOLR advantage of minimizing cable
movements this measurement should be done at the end after all calibration standards
and the DUT have been measured. If the VNA firmware or the software permit an error
correction of an earlier measurement it can also be done at the beginning before all
other measurements start.
After an SOLT calibration followed by a two-port measurement of a passive, recip-
rocal DUT the following test can be carried out: the measurement data of the DUT
can be used to perform an SOLR calibration. The SOLR error correction can then be
applied to the raw measurement of the flush thru, which has already been measured
during the SOLT calibration. The result should again provide the ideal values, allowing
for measurement uncertainties.
The verification measurement of a flush thru is a highly quantitative test that helps to
identify problems related to the characterization data for the highly reflective standards
that were used during VNA calibration.

E.2 Quantitative verification criteria


As pointed out in 4.5, the visual comparison between reference and measurement data
of a verification standard is often a good starting point to recognize problems. The

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 61 –
quantitative methods shown below, in E.2.1 for the scalar case and E.2.2 for the mul-
tidimensional case, should be regarded as supplements that can be implemented in
software. They may be used to automate any type of conformity assessment of the
measurement data. If correlation information between measurement and reference
data of the verification standards is available, see remarks in 4.5.6, the quantitative
evaluation will support a sharper pass/fail decision.

E.2.1 Scalar case


A suitable quantitative verification criteria for a scalar parameter is based on the nor-
malized difference , also referred to as normalized error (En ), between measurement
and reference data. The pass condition is  ≤ 1 for all frequency points.  is expressed
as
ˆ
|d|
=  . (E.5)
k u dˆ

The quantity d is the frequency dependent difference between measurement and refer-
ence of a scalar quantity related to the verification standard. The scalar quantity is e.g.
the magnitude of S11 . An estimate of d is in this case

ˆ
d = Ŝc11 − Ŝr11 ,

(E.6)

with the error corrected measurement and the reference data denoted by the super-
scripts c and r, respectively.
ˆ is the standard uncertainty associated with d.
u(d) ˆ It is calculated from the individ-
c r
ual uncertainties associated with |Ŝ11 | and |Ŝ11 | and, if available, from the associated
correlation coefficient r(|Ŝc11 |, |Ŝr11 |):
  r  2   2      
u dˆ = u Ŝc11 + u Ŝr11 − 2r Ŝc11 , Ŝr11 u Ŝc11 u Ŝr11 . (E.7)

k is the coverage factor used to expand the uncertainty in the denominator of equa-
tion (E.5). The value of k determines the normalization factor and determines the sta-
tistical significance of the pass criteria. Larger k represent more relaxed conditions. A
value that is often used is k = 1.96. It expands the uncertainty associated with dˆ to a
95% coverage interval. k = 1 corresponds to a 68% coverage interval. This verification
procedure assumes that the underlying PDF associated with dˆ is Gaussian.
Equations (E.6) and (E.7) are shown for the magnitude of S11 . Analog expressions
can be used for the phase of S11 or any other scalar component of an S-parameter. In
case of the phase of the S-parameter it is advised to use the unwrapped phase to avoid
problems related to the cyclicality of the phase.

E.2.2 Multivariate case


The multidimensional verification procedure uses multivariate statistics. It is beyond the
scope of this guide to provide an introduction to this topic and the reader is referred to
text books on the subject, e.g. [60], and to [5].

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 62 –
A straightforward generalization of the scalar criterion (E.5) is expressed by the
matrix equation r   
1 −1
= d̂ u d̂ d̂0 (E.8)
k
with the prime denoting the transposed. For a visual interpretation of this expression
refer to [61]. In the scalar case limit (E.8) equals (E.5). The pass criterion is as well
 ≤ 1.
d is a row vector, which contains the components of the difference between mea-
surement and reference data, e.g.
 h i h i
d̂ = Re Ŝc11 − Ŝr11 Im Ŝc11 − Ŝr11 , (E.9)

with superscripts c and r denoting error corrected measurement and reference, respec-
tively.
u(d̂) is the uncertainty (covariance) matrix associated with d̂. It is expressed as
         
u d̂ = u Ŝc11 + u Ŝr11 − u Ŝc11 , Ŝr11 − u Ŝr11 , Ŝc11 (E.10)

with u(Ŝc11 ) and u(Ŝr11 ) according to equations (C.1) and (C.2) and
  h i h i  h i h i 
  u Re Ŝ c , Re Ŝr u Re Ŝ c , Im Ŝr
11 11 11
u Ŝc11 , Ŝr11 =   h i h i  h i h 11 i  (E.11)
u Im Ŝc11 , Re Ŝr11 u Im Ŝc11 , Im Ŝr11
  h i h i  h i h i 
  u Re Ŝr , Re Ŝc u Re Ŝ r , Im Ŝc
11 11 11
u Ŝr11 , Ŝc11 =   h i h i  h i h 11 i . (E.12)
u Im Ŝr11 , Re Ŝc11 u Im Ŝr11 , Im Ŝc11

The elements in the matrices represent covariances, which are generally expressed as
u(â1 , â2 ) = r(â1 , â2 ) u(â1 ) u(â2 ) with r(â1 , â2 ) denoting the correlation coefficient and
u(â1 ) and u(â2 ) denoting standard uncertainties associated with estimates of a1 and
a2 , respectively. (E.11) and (E.12) are zero unless there is correlation between Ŝc11 and
Ŝr11 .
The coverage factor k plays the same role as in the scalar case, but for 95% cover-
age probability in the two-dimensional case the value k = 2.45 should be used, see [5].
For k = 1 the coverage probability is 39%. Again, this assumes that the probability den-
sity function associated with the measurement uncertainty of d̂ is a bivariate Gaussian
one.
Equations (E.9), (E.10), (E.11) and (E.12) are valid for S11 . Analog expressions can
be used for the other S-parameters. Instead of using real and imaginary components
the same formalism can be applied for magnitude and phase, but this should only be
done in cases where the phase is well defined, see remarks in C.1.
The same formalism can be applied to dimensions larger than two, by e.g. using all
four S-parameters that have been measured for a two-port device. This corresponds to
an eight-dimensional evaluation and the factor k needs to be adjusted accordingly.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 63 –
E.2.3 Examples of quantitative verification
Applications of the methods presented in E.2.1 and E.2.2 are shown as examples in
figures E.4 and E.5.
Figure E.4 shows the result of a verification measurement of a T-checker, see
E.1.2.5. Reference data (solid red lines) are compared with measurement data (dashed
blue lines) for the magnitude, |S11 |, and phase, arg(S11 ), of the reflection coefficient.
The three lines in each case indicate the data (center line) bounded by the interval of
the standard uncertainty. The phase is normalized to the reference data for better visual
comparison.
(|S11 |) and (arg(S11 )) are scalar normalized differences according to equation
(E.5) for the magnitude and phase of S11 , respectively. The solid line is for k = 1
and the dashed line is for k = 1.96. The solid horizontal line indicates the threshold
value of  = 1. The graph at the bottom ((S11 )) applies the multidimensional criterion
(E.8) to the vector S11 = (|S11 | arg(S11 )) with k = 1 (solid line) and k = 2.45 (dashed
line).
Figure E.5 shows the same analysis as figure E.4 but for a matched load. Instead
of magnitude and phase the real and imaginary components, Re[S11 ] and Im[S11 ], are
evaluated. In both cases, the measurement data are normalized to the reference data.
The multivariate criterion requires a condition to be met in each scalar component
in order to pass, i.e. a “bad” result in one component can’t be compensated for with a
“very good” result in another component. A failure in a single component is sufficient
to create a multivariate fail condition. The multivariate criterion is therefore a suitable
check for overall acceptance. In case of rejection it might be useful to apply the scalar
criterion to each scalar component to identify the problem.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 64 –
0.4
Reference data
Measurement data
0.35
|S11 |

0.3

0.25

0.2
4
k=1
ǫ (|S11 |)

k = 1.96
2

0.1
Reference data
Measurement data
0.05
arg (S11 )

−0.05

−0.1
4
ǫ (arg (S11 ))

k=1
k = 1.96
2

4
k=1
ǫ (S11 )

k = 2.45
2

0
10 20 30 40 50 60
Frequency /GHz
Figure E.4: Quantitative verification criteria applied to the verification measurement of a
T-checker. Detailed explanation can be found in the text.
EURAMET Calibration Guide No. 12
Version 3.0 (03/2018) – 65 –
0.03
Reference data
Measurement data
0.02

0.01
Re [S11 ]

−0.01

−0.02
4
ǫ (Re [S11 ])

k=1
k = 1.96
2

0.02

0.01

0
Im [S11 ]

−0.01

−0.02 Reference data


Measurement data
−0.03
4
ǫ (Im [S11 ])

k=1
k = 1.96
2

4
k=1
ǫ (S11 )

k = 2.45
2

0
10 20 30 40 50 60
Frequency /GHz
Figure E.5: Quantitative verification criteria applied to the verification measurement of a
matched load. Detailed explanation can be found in the text.
EURAMET Calibration Guide No. 12
Version 3.0 (03/2018) – 66 –
F VNA measurement models
F.1 Introduction
The propagation of input measurement uncertainties to the final result is based on VNA
measurement models. Such models are not unique, i.e. different levels of refinement
are possible. A VNA measurement model incorporates the same error model used
during the calibration of the VNA, see section 3, and adds the additional quantities
discussed in 5.3, which are influencing the measurements.
Sections F.2, F.3 and F.4 discuss rigorous models, covering the entire VNA mea-
surement process. Section F.5 introduces a simplified residual model, which is used by
the Ripple Method. Section F.6 briefly discusses the influence quantities in the models
with further references to the characterization of associated uncertainties in appendix
G.

F.2 One-port measurement model


Figure F.1, shows the flow graph of a one-port measurement model based on the error
model shown in figure D.1. The terms denoted by Eij are the error coefficients, which
are determined during the VNA calibration. For the names associated with the error
coefficients refer to table D.1. The additional terms are listed in table F.1 and further
explained in F.6.
From the flow graph it is possible to derive the actual measurement equations, which
represent both, the VNA calibration and the error corrected measurement of the DUT.
Analytical evaluation of figure F.1 leads to an expression for the reflection coefficient
indicated by the VNA Sm 11 = b1 /a1 :

k2 C10 C01 E001 S11


 
m 0 0
S11 = NL + NH L E00 + kC00 E01 + (F.1)
1 − (C11 + kC01 C10 E011 ) S11
with
E000 =E00 + D00
E001 =E01 D01
E011 =E11 + D11
1
k= .
1 − E011 C00
Equation (F.1) can be used to estimate the error coefficients in the VNA calibration
process. Measuring three calibration standards leads again to a set of equations, which
can be solved for the error coefficients.
Solving (F.1) for S11 leads to
Sm11 −NL
− E000 − kC00 E001
S11 = NHmL  . (F.2)
S −N
(C11 + kC01 C10 E011 ) 11 NH L
L
− E000 − kC00 E001 + k2 C10 C01 E001

Equation (F.2) can be used to determine an estimate of the reflection coefficient of the
DUT by substituting the error coefficients determined in the calibration process.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 67 –
1 1 C10
a1

E00 + D00
E11 + D11 C00 C11 S11

NH · L E01 · D01 C01


b1

NL
Figure F.1: Signal flow graph of one-port VNA measurement model. The model repre-
sents an extension of the one-port VNA error model with additional terms influencing the
measurements.

Table F.1: List of additional influence quantities in the one-port measurement model
shown in figure F.1.

Symbol Description

NL Noise floor
NH Trace noise
L Non-linearity
D00 Drift of directivity
D01 Drift of reflection tracking
D11 Drift of source match
C00 C11 Reflection of cable and connector
C01 C10 Transmission of cable and connector

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 68 –
Equations (F.1) and (F.2) are extensions of (D.1) and (D.2) by including additional
quantities influencing the measurements. Together they represent a measurement
model of the complete one-port measurement process. This model can be used to
propagate measurement uncertainties associated with the estimates of all quantities
to the error-corrected S-parameters of the DUT. These uncertainties are discussed in
detail in section 5 and explicit procedures for their quantification are given in appendix
G.

F.3 Two-port measurement models


Figures F.2 and F.3 show flow graphs representing the two-port measurement models
based on the error models shown in D.2. For the names associated with the error
coefficients refer to tables D.2 and D.3. The additional terms are principally the same
as for the one-port model, see table F.1, except that additional drift terms have to be
considered for transmission tracking and load match in the 10-term model and for the
switch term in the 7-term model. Obviously all influence quantities occur twice, once for
each port, and the indices are chosen accordingly. Further details on these influence
quantities are given in F.6.
From the flow graph it is possible to determine the actual measurement equations,
representing both, VNA calibration and VNA error correction, in the same way as it is
done for the one-port case in F.2. The equations become significantly more elaborate
and are therefore not shown here. In practice the equations can be derived by using
matrix flow graph reduction [16] and symbolic mathematical computation software.
The equations represent a measurement model that can be used to propagate un-
certainties associated with estimates of all influence quantities in a two-port measure-
ment to the error corrected S-parameter of the DUT.

F.4 N-port measurement model


The same concepts can be used for an N-port model as for the one-port and two-port
models described in F.2 and F.3. An example of a generalized N-port version of a
measurement model can be found in [59].

F.5 Measurement model based on residual error coefficients


The one-port measurement model based on residual error coefficients, in short a “resid-
ual measurement model”, is shown in figure F.4. It is equivalent to the one-port mea-
surement model shown in figure F.1, but it assumes that the measurement of the DUT
is taken with an error corrected VNA. Hence, the error coefficients Eij are replaced by
deviations from the ideal values. These deviations are called residual error coefficients
ˆ 00 = δE
δEij , with estimates δE ˆ 11 = δE
ˆ 01 = 0. The indicated reflection coefficient Sm
11
is replaced by the error corrected reflection coefficient Sc11 . Because all other terms
in the model have estimates of 0 or 1 too, it is obvious that Ŝc11 = Ŝ11 . The residual
measurement model is an approximation, because in reality a VNA is not measuring
error corrected S-parameters. Validity and applicability of this model for the purpose of
uncertainty evaluation are based on assumptions [62].

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 69 –
E30

1 1 C10 S21
a0

EF00 + DF00
EF11 + DF11 C00 C11 S11 S22
NH1 · L1 EF01 · DF01 C01 S12
b1

NL1 E30 NL2

C32 EF32 · DF32 NH2 · L2


b2

C22 C33 EF22 + DF22


C23

Figure F.2: Signal flow graph of two-port VNA measurement model. The model repre-
sents an extension of the ten-term forward VNA error model with additional terms influ-
encing the measurements. Due to the limitations in page width the graph is separated in
two parts, connected at the dashed lines. Accordingly, the measurement model for the
reverse mode is identically based on the ten-term reverse VNA error model.

E30

1 1 C10 S21
a1

E00 + D00
W00 + V00 E11 + D11 C00 C11 S11

NH1 · L1 E01 · D01 C01 S12


b1
E03
NL1

Figure F.3: Signal flow graph of two-port VNA measurement model (only the port 1 side
is shown). Port 2 is symmetrical with the dashed line, except that both reflection track-
ing terms have to be considered at the port 2 side, see table D.3 and figure D.5. The
model represents an extension of the seven-term VNA error model with additional terms
influencing the measurements.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 70 –
1 1 C10

δE00 + D00
Sc11 δE11 + D11 C00 C11 S11

NH · L (1 + δE01 ) · D01 C01

NL
Figure F.4: Signal flow graph of residual one-port VNA measurement model. The model
is equivalent to the one-port measurement model shown in figure F.1, except that the
error coefficients Eij are replaced by residual error coefficients δEij and the indicated
reflection coefficient Sm c
11 is replaced by the error corrected reflection coefficient S11 .

Using the same approach as for the one-port case, residual two-port measurement
models can be formulated, e.g. the corresponding forward model would be based on
figure F.2.
The residual measurement model has the advantage that it leads to relatively simple
equations for the propagation of measurement uncertainties. The Ripple Method, which
is discussed in appendix H, is based on the residual measurement model with some
further simplifications. The resulting equations for linear uncertainty propagation are
shown in H.4.
It is known, however, from experience that in transmission measurements the load
match has an amplifying effect on the uncertainty, which is often not negligible. Such an
effect is simply not in the scope of the residual measurement model and it is necessary
to take additional measures to avoid an underestimation of the uncertainty, see H.4.2.1.
Use of the full model, as done for the rigorous uncertainty evaluation, doesn’t have that
problem.

F.6 Uncertainty contributions


In the following the different terms of the particular models shown in figures F.1, F.2 and
F.3 are briefly discussed and put into relation with the characterization of associated
uncertainties in 5.3. The nomenclature refers to the model in figure F.3 but is easily
transferable to the other models.
The terms denoted by Eij are the error coefficients. Estimates Êij are determined
during VNA calibration. The measurement model includes potential drift of these terms,
denoted by Dij except for the cross talk (E30 and E03 ), which is generally very small.
Best estimates D̂ij are 0 (when additive) and 1 (when multiplicative) with associated
uncertainties as determined in G.3. The term W00 denotes the switch term with associ-
ated drift V00 and V̂00 = 0. W00 is shown with a dotted line, as it is only present when
the source is active at port 2. For four-receiver VNAs, which have the capability to read
out all receivers simultaneously, it can be determined directly for each measurement

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 71 –
and is not part of the normal calibration process.
The terms denoted by Cij represent effects due to the movement of test port cables
and due to connector repeatability. The relation to the characterization in G.4 and G.5
can be established through a simple sub-model as shown in figure F.5. The terms
CAT and CAR are related to cable movement and have best estimates of 1 and 0,
respectively, with associated uncertainties as specified in G.4. The term COR is related
to connector repeatability and has a best estimate of 0 with an associated uncertainty as
specified in G.5. Cascading the two 2-ports in figure F.5 leads to the following relations
(shown here just for port 1):

CA2T COR
C00 = CAR +
1 − COR CAR
CAR
C11 = COR +
1 − COR CAR
CAT
C10 = C01 = .
1 − COR CAR
These expressions simplify considerably if used for linear uncertainty propagation with
best estimates of unity and zero.

C00 = C11 = CAR + COR


C10 = C01 = CAT .

The term NLi denotes noise floor with N̂Li = 0. The terms NHi and Li denote
trace noise and non-linearity, respectively, with N̂Hi = 1 and L̂i = 1. The associated
uncertainties can be determined using the procedures in G.1 and G.2, respectively.

Cable Connector
CAT 1

CAR CAR
COR COR

CAT 1

Figure F.5: Signal flow graph representing a simple measurement sub-model for cable
movement and connector repeatability at port one.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 72 –
G Characterization procedures of uncertainty contributions
The following step by step characterization procedures complement the discussion on
uncertainty contributions in section 5. They are specifically chosen to be used with the
measurement models in appendix F. The procedures below are examples, i.e. other
ways of characterization are possible.
Fundamentally, procedures are shown for two types of influencing quantities, de-
pending on how they act in the measurement model, additive quantities with an estimate
of 0 and multiplicative quantities with an estimate of 1. Additive quantities are noise
floor, drift related to directivity and match terms, cable and connector reflections. The
associated uncertainties are characterized in Cartesian coordinates (real and imaginary
components) in linear units. Multiplicative quantities are trace noise, non-linearity, drift
related to the transmission error coefficients and cable transmission. The associated
uncertainties are characterized in polar coordinates (magnitude and phase) in linear
units.

G.1 Noise floor and trace noise


This procedure is applicable to a two port VNA. It is suitable to determine measurement
uncertainties associated with the best estimates of the terms NLi (noise floor) and NHi
(trace noise) in the measurement models shown in appendix F. Best estimates of NLi
and NHi are N̂Li = 0 and N̂Hi = 1.

a) VNA settings: leave the VNA uncalibrated and set the test port power, IF band-
width and averaging to the values that are being used for measurements. Set the
sweep type to CW (continuous wave).

b) Connect two shorts or opens to port 1 and to port 2. Choose the standard, which
shows a larger reflection coefficient. Wait a few minutes.

c) Collect at least a few hundred measurement data points of Sm m


11 and S22 (for the
m m
calculation of the trace noise) and S21 and S12 (for the calculation of the noise
floor) for each frequency point.

d) Calculate the standard deviations of the real and imaginary components of Sm 21 .


m m
Take the larger of both: max(s(Re[S21 ]) , s(Im[S21 ])) with s(·) according to equation
(C.7).

e) Define a reasonable envelope that encloses the calculated values over the whole
frequency range. Typically, the whole frequency range is divided into frequency
bands, each with a constant value that encloses the calculated values. The values
can be taken as the standard uncertainties associated with the real and imaginary
components of N̂L2 .

f) Calculate the standard deviation of the normalized magnitude of Sm 11 :


s(|Sm11 /m(S m ) |) with m(·) and s(·) according to equations (C.9) and (C.7), respec-
11
tively.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 73 –
g) Calculate the standard deviation of the normalized phase of Sm 11 :
m m
s(arg[S11 /m(S11 )]) with m(·) and s(·) according to equations (C.9) and (C.7), re-
spectively.

h) For the values defined in step f) and g), define envelopes according to step e) and
assign the values as uncertainties associated with magnitude and phase of N̂H1 .

i) Repeat steps d) to h) for Sm m


12 and S22 to determine uncertainties associated with
N̂L1 and N̂H2 , respectively. Normally, this should give very similar results and the
same uncertainties can be assigned to both ports by just taking the larger values
for both.

Figures G.1, G.2, G.3, G.4, G.5 and G.6 are illustrating the above procedure.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 74 –
−4
x 10
1.5

0.5
21 ]
Re [Sm

−0.5

−1

−1.5
0 10 20 30 40 50 60 70
Frequency /GHz

Figure G.1: Characterization of VNA noise floor according to the procedure in G.1. The
blue data points are measurements of Sm 21 according to step c) of the procedure. Only the
real component of the S-parameter is shown. The imaginary component is not shown
here, but gives very similar results. 801 measurements were made per frequency point.
The black circles with bars indicate mean and standard deviation s(Re [Sm 21 ]) of the mea-
sured sample at each frequency point, according to step d). The orange line indicates a
possible envelope, according to step e). Note that the lowest frequency in this example
is 1 GHz. At lower frequencies significantly larger noise floor is conceivable, see figure
G.2.

−3
x 10
6

2
21 ]
Re [Sm

−2

−4

−6
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Frequency /GHz

Figure G.2: The same as figure G.1, but for frequencies below 0.1 GHz. Note the sig-
nificantly enlarged noise level towards low frequencies due to the reduced efficiency of
couplers.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 75 –
−4
x 10
2

1.5

1
11 /m (S11 )| − 1

0.5

0
m

−0.5
|Sm

−1

−1.5

−2

−2.5
0 10 20 30 40 50 60 70
Frequency /GHz

Figure G.3: Characterization of VNA trace noise magnitude according to the procedure
in G.1. The blue data points are normalized measurements of Sm 11 according to step c) of
the procedure. 801 measurements were made per frequency point. The black circles with
bars indicate mean and standard deviation of the displayed sample at each frequency
point, according to step f). The orange line indicates a possible envelope, according to
step h). Larger values of trace noise can occur for values at lower frequencies, see figure
G.4.

−3
x 10
2.5

1.5

1
11 /m (S11 )| − 1

0.5
m

−0.5
|Sm

−1

−1.5

−2

−2.5
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Frequency /GHz

Figure G.4: The same as figure G.3, but for frequencies below 0.1 GHz. Note the sig-
nificantly enlarged noise level towards low frequencies due to the reduced efficiency of
couplers.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 76 –
0.04

0.03

0.02

11 /m (S11 )) /

0.01
m

0
arg (Sm

−0.01

−0.02

−0.03

−0.04
0 10 20 30 40 50 60 70
Frequency /GHz

Figure G.5: Characterization of VNA trace noise phase according to the procedure in G.1.
The blue data points are normalized measurements of Sm 11 according to step c) of the
procedure. 801 measurements were made per frequency point. The black circles with
bars indicate mean and standard deviation of the displayed sample at each frequency
point, according to step g). The orange line indicates a possible envelope, according to
step h). Larger values of trace noise can occur for values at lower frequencies, see figure
G.6.

0.15

0.1

0.05

11 /m (S11 )) /
m

0
arg (Sm

−0.05

−0.1

−0.15

−0.2
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Frequency /GHz

Figure G.6: The same as figure G.5, but for frequencies below 0.1 GHz. Note the sig-
nificantly enlarged noise level towards low frequencies due to the reduced efficiency of
couplers.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 77 –
G.2 VNA non-linearity
This procedure is applicable to a two port VNA and uses an automatic step attenuation
device. Manually operated step attenuation devices usually show worse repeatability.
The procedure is suitable to determine uncertainties associated with the best estimate
of the non-linearity term Li in the measurement models given in appendix F. The best
estimate of Li is L̂i = 1

a) Set the test port power of the VNA to the value that is being used for the mea-
surements.

b) Set the IF bandwidth of the VNA to 10 Hz and the average factor to 1.

c) Perform a full two port calibration of the VNA. The SOLR calibration is preferred
in order to avoid unnecessary cable movement.

d) Connect a characterized step attenuation device with a maximum attenuation of


at least 60 dB. A device with 10 dB steps is sufficient for this type of assessment.
The characterized incremental step width of the attenuation device is denoted by
Aref (j) with j = 1, 2, 3, ... indicating the attenuation step.

e) Measure all states of the step attenuation device over the whole frequency range
and perform a VNA error correction to obtain estimates, Ŝ21 (j) and Ŝ12 (j), of
the S-parameters of the attenuation device for each state j = 0, 1, 2, 3, ... (j = 0
corresponds to the through state and represents the residual attenuation of the
device).

f) Compute estimates of the incremental attenuations Â1 (j) = |Ŝ21 (j)| − |Ŝ21 (0)| of
the error corrected measurements. |Ŝ21 (j)| and |Ŝ21 (0)| need to be in logarithmic
units. Repeat this process for S12 to calculate Â2 (j).

g) Take the larger of the two differences between measured and characterized incre-
mental attenuations: dj = max(|Â1 (j) − Aref (j)|, |Â2 (j) − Aref (j)|).

h) Define a reasonable envelope that encloses the calculated differences dj over the
whole attenuation range. Since the frequency dependence is generally weak, this
envelope might enclose any frequency dependence.

i) Transform dj to linear units using dj,lin ' 0.12 dj and use dj,lin as uncertainties
u(|L̂i |) and u(arg(L̂i )).

The procedure above is limited by the uncertainty associated with the attenuation
steps. These uncertainties should be taken into account, if significant. The stated non-
linearities can’t be smaller than the uncertainty associated with the attenuation steps.
Figures G.7, G.8 and G.9 illustrate the procedure and provide additional information.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 78 –
2

1.5
Aˆ1 − Aref /dB

0.5

−0.5
−90 −80 −70 −60 −50 −40 −30 −20 −10
Aref /dB

Figure G.7: Characterization of VNA non-linearity according to procedure G.2. The dif-
ference between a measurement and the reference value of incremental attenuation is
shown for the entire attenuation range down to -90 dB at 1 GHz. This corresponds to
step g) of the procedure, although only showing data for S21 . The different colors indi-
cate four repetitions of the same measurement sequence. The large deviation from 0 and
enhanced scatter below -50 dB are due to the influence of noise floor. Therefore, for the
characterization of non-linearity only data above -60 dB should be considered, see figure
G.8

0.05

0.04

0.03

0.02
Aˆ1 − Aref /dB

0.01

−0.01

−0.02

−0.03

−0.04

−0.05
−55 −50 −45 −40 −35 −30 −25 −20 −15 −10 −5
Aref /dB

Figure G.8: The same as figure G.7. Only data above -60 dB should be used to char-
acterize VNA non-linearity. Uncertainty intervals are shown as well. They need to be
considered, when defining the envelope according to step h) of the procedure.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 79 –
0.05

0.04

0.03

0.02
Aˆ1 − Aref /dB

0.01

−0.01

−0.02

−0.03

−0.04

−0.05
−55 −50 −45 −40 −35 −30 −25 −20 −15 −10 −5
Aref /dB

Figure G.9: The same as figure G.7. A single measurement sequence at three different
frequencies, 1 GHz (blue), 10 GHz (green) and 18 GHz (red) is shown. The data indicates
that the frequency dependence of non-linearity is weak. The orange line indicates a
possible envelope, according to step h).

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 80 –
G.3 VNA drift
These procedures are valid for a two port VNA. They determine uncertainties associ-
ated with estimates of the drift of VNA error coefficients, as represented by the Dij and
Vij terms in the VNA measurement models in appendix F. The characterization has
to be performed under normal operating and environmental conditions, taking typical
variations into account. It is necessary to repeat the characterization, if environmental
conditions change.
Two methods are given. The first uses an electronic calibration unit (ECU), the sec-
ond can be carried out without ECU. Recording of ambient temperature in parallel might
be useful for the interpretation of the data.

Method with ECU:


This method can characterize the drift of each individual error coefficient.

a) Connect ECU to the VNA.

b) Measure the states of the ECU repeatedly over a period of 24 hours with an inter-
val of 15 minutes. This necessitates computerized control of the ECU switching
states (refer to VNA manual, ECU manual or contact manufacturer).

c) Use each measurement set to determine estimates of the error coefficients of the
VNA as a function of time Êij (t), starting at t = 0.

d) Determine the drift of the error coefficients related to directivity and match and the
switch terms with respect to the first measurement at t = 0 with ∆Eij (t) = Êij (t) −
Êij (t = 0). Calculate the maximum deviation as max(| Re[∆Eij (t)]|, | Im[∆Eij (t)]|).
Define a reasonable envelope that encloses the maximum deviation over the 24
hour time span. This could e.g. be a linear function of time and/or frequency.
Take the values of this envelope as standard uncertainties associated with the
estimates of both the real and imaginary components of the drift terms. It is im-
portant that these drift functions are defined rather generously to enclose possible
variations in drift behavior.

e) Determine the drift of the error coefficients related to transmission with ∆Eij (t) =
Êij (t)/Êij (t = 0). Calculate the deviations in magnitude as |∆Eij (t)| − 1 and in
phase as | arg(∆Eij (t))| Define a reasonable envelope that encloses the devia-
tions over the 24 hour time span. This could e.g. be a linear function of time
and/or frequency. Take the values of these envelopes as standard uncertainties
associated with the estimates of magnitude and phase of the drift terms. It is im-
portant that these drift functions are defined rather generously to enclose possible
variations in drift behavior.

Method using mechanical calibration standards:

a) Perform a two-port calibration of the VNA.

b) Establish a flush thru by connecting both test port cables.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 81 –
c) Measure the full set of S-parameters repeatedly over a period of 24 hours with an
interval of 15 minutes and perform VNA error corrections on each set to obtain
estimates Ŝij (t) starting at t = 0.

d) Determine the drift of S11 and S22 with respect to the first measurement at t = 0
with ∆Sii (t) = Ŝii (t) − Ŝii (t = 0). Calculate the maximum deviation as
max(| Re[∆S11 (t)]|, | Im[∆S11 (t)]|, | Re[∆S22 (t)]|, | Im[∆S22 (t)]|). Define a reason-
able envelope that encloses the maximum deviation over the 24 hour time span.
This could e.g. be a linear function of time and/or frequency. Take the values of
this envelope as standard uncertainties associated with the estimates of both real
and imaginary components of the drift terms related to the error coefficients direc-
tivity, match and switch terms. It is important that these drift functions are defined
rather generously to enclose possible variations in drift behavior.

e) Determine the drift of S21 and S12 with respect to the first measurement at t = 0
with ∆Sij (t) = Ŝij (t)/Ŝij (t = 0). Calculate the maximum deviations in magnitude
and phase as max(|∆S21 (t)|−1, |∆S12 (t)|−1) and max(| arg[∆S21 (t)]|, | arg[∆S12 (t)]|)
Define a reasonable envelope that encloses the maximum deviations over the 24
hour time span. This could e.g. be a linear function of time and/or frequency.
Take the values of these envelopes as standard uncertainties associated with the
estimates of magnitude and phase of the drift terms associated with the error
coefficients related to transmission. It is important that these drift functions are
defined rather generously to enclose possible variations in drift behavior.

Figures G.10, G.11 and G.12 illustrate the procedure without ECU. At very low fre-
quency the enhanced noise floor due to inefficient couplers is visible again, in particular,
in figures G.10 and G.11. This effect should already be taken into account in the char-
acterization of the VNA noise floor, described in G.1, and can be ignored in the drift
characterization.
Drift is affected by changes in the environment. After performing the VNA calibra-
tion, the operator connects the flush thru, starts the automated measurement sequence
and then leaves the setup alone. The effect of this handling at the start of the procedure
can be seen in figure G.12 when the phase of the transmission coefficient shows the
largest deviation at the beginning of the sample of observations. To a lesser extent it
can be also observed for the magnitude (figure G.11). Later in a quiet environment, the
changes become more gradual.
It is important to understand that the procedure above characterizes drift related to
the entire setup, consisting of VNA and test port cable. Using another test port cable
might provide different results.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 82 –
−3
x 10

4
Ŝii (t) − Ŝii (t = 0)

−2

−4

−6

0 10 20 30 40 50 60
Frequency /GHz

Figure G.10: Characterization of VNA drift with a flush thru measurement haccording
i to
the second procedure in G.3. The traces are repeated measurements of Re Ŝ11 (black),
h i h i h i
Im Ŝ11 (blue), Re Ŝ22 (green), Im Ŝ22 normalized to the first measurements, accord-
ing to step d) of the procedure. The measurements were done every 15 minutes for a
duration of 24 hours. The orange line indicates a possible envelope, according to step
d).

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 83 –
0.01

0.008

0.006
-Ŝij (t)/Ŝij (t = 0)- − 1

0.004
-
-

0.002

−0.002

−0.004
-
-

−0.006

−0.008

−0.01
0 10 20 30 40 50 60
Frequency /GHz

Figure G.11: Characterization of VNA drift with a flush thru measurement according
to
the second procedure in G.3. The traces are repeated measurements of Ŝ21 (green) and


Ŝ12 (blue) normalized to the first measurements, according to step e) of the procedure.

The measurements were done every 15 minutes for a duration of 24 hours. The orange
line indicates a possible envelope, according to step e).

1.5

1
arg Ŝij (t)/Ŝij (t = 0) /◦
2

0.5

−0.5
1

−1

−1.5
0 10 20 30 40 50 60
Frequency /GHz

Figure G.12: Characterization of VNA drift with a flush thru measurement according
 to
the second procedure in G.3. The traces are repeated measurements of arg Ŝ21 (green)
 
and arg Ŝ12 (blue) normalized to the first measurements, according to step e) of the
procedure. The orange line indicates a possible envelope. The measurements were
done every 15 minutes for a duration of 24 hours.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 84 –
G.4 Test port cable stability
This procedure aims at a relatively simple characterization of uncertainties associated
with estimates of the terms CAR (cable reflection) and CAT (cable transmission) in
figure F.5. Best estimates are CAˆ R = 0 and CA ˆ T = 1. More sophisticated character-
ization procedures have been developed as well [63]. See also 8.5 for some practical
hints on optimizing the setup of test port cables.
a) Perform a one-port calibration of the VNA at the VNA port.

b) Connect the test port cable to the VNA port. Wait for at least 30 minutes to achieve
thermal equilibrium. To avoid the waiting time see figures G.15 and G.16 for an
alternative setup.

c) Connect a short to the end of the test port cable.

d) Hold the cable in a straight line and take a first measurement (n = 1) of magnitude
and phase.

e) Perform a gradual bend to a defined position. The defined position might be a


90 deg bend in the horizontal plane defined by the test port connectors of the
VNA (8.5.1). If a subsequent measurement requires a cable bend leaving the
horizontal plane, i.e. a vertical bend, this needs to be executed as well, because
the behavior might be different due to the natural bending of the test port cable.
In any case the characterization should enclose any movements that might occur
in a subsequent measurement.

f) Take several measurements n = 2, 3, ... of magnitude and phase during the grad-
ual bend. Record the maximum deviation in magnitude and phase of the reflec-
tion coefficient with respect to measurement n = 1: ∆r = maxn (||Ŝ11 (n)/Ŝ11 (n =
1)| − 1|) and ∆φ = maxn (arg(Ŝ11 (n)/Ŝ11 (n = 1))). Define a generous envelope
over frequency, enclosing the maximum deviations.

g) Use the envelope values determined in the previous step as uncertainties that are
ˆ T |) =
associated with the estimate of the term CAT in figure F.5 leading to u(|CA
ˆ
∆r/2 and u(arg(CAT )) = ∆φ/2. The division by 2 for magnitude and phase is
applied, because the characterization with a short evaluates the stability of the
cable for a signal that propagates back and forth in the cable. The uncertainty
terms, however, are related to one-path effects.

h) For the term CAR in figure F.5 steps c) to g) have to be repeated with a matched
load connected to the VNA cable instead of a short. In this case, the measured
magnitude of S11 is small and hence the phase has minor significance. Thus,
an evaluation in real and imaginary components is preferable, leading to ∆x =
maxn (Re[Ŝ11 (n) − Ŝ11 (n = 1)]) and ∆y = maxn (Im[Ŝ11 (n) − Ŝ11 (n = 1)]). The
uncertainties assigned to the real and imaginary parts of an estimate of CAR are
ˆ R ) = u(Re[CA
u(CA ˆ R ]) = u(Im[CA ˆ R ]) = max(∆x, ∆y).

Figures G.13 and G.14 illustrate the characterization of the cable with respect to
transmission stability. The corresponding measurement setups are shown in figures

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 85 –
G.15, for the start position of the cable, and G.16, for the end position of the cable (after
gradual bend). The starting position is in this case not the straight cable. Instead the
positional variation shown might be typical for two port measurements of a small-sized
DUT. The characterization is viable as long as cable movements are kept within these
limits. A clamping system, see 8.5.1, is helpful to keep control of cable positions.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 86 –
0.03

0.02
-Ŝ11 (n)/Ŝ11 (n = 1)- − 1

0.01
-
-

−0.01
-
-

−0.02

−0.03
0 10 20 30 40 50 60
Frequency /GHz

Figure G.13: Characterization of transmission magnitude stability of VNA test port cable
according to the procedure in G.4. The traces are repeated normalized measurements,
corresponding to step f) of the procedure, with the short connected to the end of the test
port cable. The orange line indicates a possible envelope. The measurement setup is
shown in figures G.15 (initial position of cable for measurement n = 1) and G.16 (final
position after gradual bend of cable).

4
arg Ŝ11 (n)/Ŝ11 (n = 1) /◦
2

−2
1

−4

−6

0 10 20 30 40 50 60
Frequency /GHz

Figure G.14: The same as figure G.13 but now for the phase.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 87 –
Figure G.15: Starting position (measurement n = 1) for the characterization of a VNA
test port cable. In this setup the VNA has been calibrated at the end of the clamped test
port cable on the left. The cable under test is the second cable, which is attached to the
clamped test port. This way the cable under test is isolated from the warm VNA port and
the waiting time of step b) is not necessary. In this setup it is important to keep the test
port cable on the left fixed with clamps.

Figure G.16: The same as figure G.15, but showing the end position for the characteriza-
tion of a VNA test port cable.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 88 –
G.5 Connector repeatability
This procedure can be used for characterization of uncertainties associated with the
estimate of the term COR (connector reflection) in figure F.5. The estimate of COR is
ˆ R = 0. The procedure aims at characterizing the typical connector repeatability of a
CO
connector family. It therefore needs to be repeated for each connector family.

a) Perform a one-port calibration.

b) Connect a short and measure Sm


11 . Perform a VNA error correction to determine
an estimate Ŝ11

c) Repeat the previous step by re-connecting the short under different azimutal posi-
tions, at least 16 times to obtain a sample of estimates: (Ŝ11 (1) Ŝ11 (2) Ŝ11 (3) . . . )

d) Calculate the difference between maximum and minimum values of the sample in
the real component of Ŝ11 : ∆x = maxi (Re[Ŝ11 (i)]) − mini (Re[Ŝ11 (i)])

e) Calculate the difference between maximum and minimum values of the mea-
surement sample in the imaginary component of Ŝ11 : ∆y = maxi (Im[Ŝ11 (i)]) −
mini (Im[Ŝ11 (i)]).

f) Take the larger of the two differences ∆z = max(∆x, ∆y)

g) Define a reasonable envelope that encloses ∆z/2 over the whole frequency range.
Typically, this will be done by dividing the whole frequency range into frequency
bands, each with a constant value that encloses the calculated values. This can
be taken as the standard uncertainties associated with the real and imaginary
components of CO ˆ R.

NOTE — From a statistical point of view it seems more appropriate to take the standard
deviation of the measurement sample as a measurement uncertainty, following the formalism in
C.5. instead of the interval enclosing all values. Taking the interval that encloses all values is a
more conservative approach, acknowledging the fact that this effect can strongly vary from one
connector pair to another within the same connector family.
Figure G.17 shows example data for the evaluation of connector repeatability for the
1.85 mm connector family. The data has been collected according to the procedure
above.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 89 –
−3
x 10

4
1 2

2
Ŝ11 − m Ŝ11

−2

−4

−6

0 10 20 30 40 50 60
Frequency /GHz

Figure G.17: Evaluation data for connector repeatability.


h iŜ11 has been determined
h i for
nine different connector orientations. Shown are Re Ŝ11 (blue) and Im Ŝ11 (green)
normalized to the mean. The orange line indicates a possible envelope, according to
step g) of the procedure.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 90 –
H Ripple Method
H.1 Introduction
This section describes the implementation of the Ripple Method for the evaluation of
measurement uncertainties in coaxial S-parameter measurements. The method, as
described here, is recommended only up to 26.5 GHz, for coaxial line sizes of 3.5 mm
and larger (e.g. Type-N), for reasons given in 1.2 and 7.2. The experimental part of the
determination of the residual error coefficients is identical to that in the previous version
of the guide. For the residual tracking term, which cannot be accessed experimen-
tally, a formula is given that allows determination of the associated uncertainty from the
characterization data of a data based calibration kit. Modifications are applied in the
analysis of the ripple data by specifying minimal uncertainty limits for residual directivity
and residual source match. The equations that propagate the uncertainties associated
with the error coefficients to the DUT are modified in accordance with the suggestions
in [32]. Additional uncertainty contributions are taken into account based on a more
rigorous derivation of a measurement model for the error correction process.

H.2 Uncertainties
The uncertainties derived in this context do not need to be interpreted as uncertain-
ties in the estimated magnitudes of the complex-valued quantities, as was the case
in previous versions of the guide. Instead, they can be seen as equal sized uncer-
tainties associated with the real and the imaginary components of the complex-valued
estimate of the S-parameter, assuming there is no correlation between them, e.g. for
S11 : u(Ŝ11 ) = u(Re[Ŝ11 ]) = u(Im[Ŝ11 ]). This corresponds to a circular uncertainty re-
gion in the complex plane, following the suggestions in [64]. This uncertainty can be
transformed to an uncertainty in magnitude and phase using standard uncertainty prop-
agation techniques.

H.3 Practical Preparation


The Ripple Method requires at least one beadless air-dielectric line, for each type of
connector. The nominal length of the air-dielectric line should be between 75 mm and
300 mm. Generally, the length of the air-dielectric line and the frequency range should
be chosen such that several full cycles of the ripple pattern are visible in the procedures
described in H.5.1 and H.5.2.
More discrimination, i.e. more ripples, will be obtained with a longer air-dielectric
line. A 300 mm air-dielectric line will give ripples having a period of 0.5 GHz and a
100 mm air-dielectric line will result in ripples with a period of 1.5 GHz. The line length
limits the lowest frequency at which a meaningful result can be obtained. E.g. in Type-N
it should be possible to make reasonable estimates of the residual error coefficients at
frequencies in the range of 0.5 GHz to 17.5 GHz with a frequency spacing of 250 MHz
using a 300 mm air-dielectric line. Estimates between these discrete frequency points
might be obtained through interpolation.
The physical dimensions of the air-dielectric line should be characterized and trace-
able to SI units. The relevant dimensions are the average outer diameter of the center

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 91 –
conductor and the average bore diameter of the outer conductor. In addition, it is impor-
tant to verify that the difference in length between the center and outer conductors of
each air-dielectric line is within specification [65]. This length measurement can usually
be made using a pin-depth gauge, often found in some VNA calibration kits. The ratio
of the diameters of the center and outer diameters can be used to determine the char-
acteristic impedance of the air-dielectric line. A conversion between the geometrical
dimensions and the electrical characteristic impedance, Z0 , for nominal 50 Ω lines, can
be approximated by  
do
Z0 ' 59.939 ln , (H.1)
dc
where do and dc denote the outer and center conductor diameters of the air-dielectric
line, respectively. Deviations in the geometrical dimensions can be used to estimate
the deviation in the electrical characteristic impedance of the air-dielectric line, based
on equation (H.1).

H.4 Measurement model


The Ripple Method is partly a “black box” method. The measurement model is based
on the residual measurement model shown in F.5, basically assuming that the VNA is
measuring error corrected S-parameters. Estimates of the residual error coefficients
are determined experimentally and uncertainties only need to be propagated through
the VNA error correction process. With a few additional simplifications it is possible to
obtain reasonably simple equations for the uncertainties associated with estimates of
the S-parameters of the DUT.

H.4.1 One-port equation


The one-port measurement model for reflection measurements is based on an eval-
ˆ 00 = δE
uation of figure F.4 with δE ˆ 01 = δE
ˆ 11 = 0. For simplification it is necessary
to keep the test port cable in a fixed position during measurement, see 8.5.1. Hence,
uncertainty contributions from cable movement don’t have to be considered. All ran-
dom contributions, e.g. connector repeatability, are summarized in a single additive
term RS11 . Linear uncertainty propagation leads to an expression for the uncertainty
associated with an estimate of S11 of the DUT
h  i2 h  i2 h  i2  2  2 h  i2
u Ŝ11 ˆ
= u δE00
c ˆ
+ Ŝ11 u δE01
c ˆ
+ Ŝ11 u δE11 + Ŝc11 u L̂

+
h  i2
u R̂S11 .
(H.2)

Sc11 denotes the error corrected reflection coefficient. The residual error coefficients
correspond to those listed in table D.1. The drift terms Dij are omitted in equation (H.2).
They have to be considered as a part of the uncertainties associated with the estimates
of the residual error coefficients. The term L denotes non-linearity.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 92 –
H.4.2 Two-port equations
In analogy to the one-port case, the two-port measurement model is based on the two-
port residual measurement model. For simplification, it is also necessary to fix the test
port cable of port one, hence only cable movements at port two have to be considered.
Using the same approach as the one-port case, the expressions for the uncertainties
associated with estimates of the S-parameters of the DUT are:
h  i2 h  F i2 h  F i2  2  F 2
u Ŝ11 = u δE ˆ ˆ
+ Ŝc11 u δE
ˆ
+ Ŝc11 u δE +
00 01 11
h  F i2 h  i2 h  i2
ˆ + (H.3a)
c c c c c
Ŝ21 Ŝ12 u δE 22 + Ŝ11 u L̂ + Ŝ Ŝ
21 12 u Ĉ22
h  i2
u R̂S11
h  i2 h  F i2 h  F i2 h  F i2
u Ŝ21 ˆ
= Ŝc11 Ŝc21 u δE +
c c
Ŝ ˆ + Ŝc21 u δEˆ
22 21 u δE22
Ŝ +

11 32
h  F i2 h  i2 h  i2
u δE ˆ + Ŝc21 u L̂

+ Ŝc21 u Ĉ32

+
30
h  i2 h  i2
c c
Ŝ21 Ŝ22 u Ĉ22 + u R̂S21
(H.3b)
h  i2 h  R i2 h  R i2  2  R  2 
u Ŝ22 = u δEˆ +
c
Ŝ u ˆ
δE +
c
Ŝ22 u δE ˆ +
33 22 32 22
h  R i2 h  i2 h  i2
c c ˆ
Ŝ21 Ŝ12 u δE +
c
Ŝ u L̂ +
c
Ŝ u Ĉ32 +
11 22 22
h  i2  2   2 h 
 i2 h  i2
c
Ŝ22 u Ĉ23 + Ŝc22 u Ĉ22 + u Ĉ33 + u R̂S22

(H.3c)
h  i2 h  R i2 h  R i2 h  R i2
u Ŝ12
ˆ
= Ŝc22 Ŝc12 u δE
ˆ
+ Ŝc11 Ŝc12 u δE
ˆ
+ Ŝc12 u δE +
22 11 01
h  R i2 h  i2 h  i2
u δE ˆ + Ŝc12 u L̂

+ Ŝc12 u Ĉ23

+
03
h  i2 h  i2
c c
Ŝ12 Ŝ22 u Ĉ22 + u R̂S12 .
(H.3d)
The terms Scij denote error corrected S-parameters. The residual error coefficients
δEFij of the forward model and δER ij of the reverse model correspond to those listed
in table D.2. Again, the drift terms are part of the uncertainties associated with the
estimates of the residual error coefficients, as in the one-port case. The terms Cij
denote cable movements on the port two side. The term L denotes non-linearity. The
terms RSij summarize all random effects.
H.4.2.1 Limitations of the residual measurement model
From practical experience it is known that uncertainties associated with transmission
measurement can be underestimated with equations (H.3b) and (H.3d), if the raw load
match of the VNA is significant. This is a shortcoming of the residual measurement

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 93 –
model, which doesn’t consider the actual values of the error coefficients, see F.5. There-
fore, instead of (H.3b) and (H.3d) the following equations are recommended [66]:
h  i2 h  F i2 h  F i2 h  F i2
u Ŝ21 ˆ
= Ŝc11 Ŝc21 u δE +
c c
Ŝ ˆ ˆ
+ Ŝc21 u δE
22 21 u δE22
Ŝ +

11 32
h  F i2 h  i2 h  i2
u δE ˆ + Ŝc21 u L̂

+ Ŝc21 u Ĉ32

+
30
h  i2 h  i2
c c
Ŝ21 Ŝ22 u Ĉ22 + u R̂S21 +
  2 h  R i2 h  F i2 

c F c c
Ŝ21 Ê22 1 − Ŝ21 Ŝ12 u δE ˆ + u δE ˆ
33 11

(H.4a)
h  i2 h  R i2 h  R i2 h  R i2
u Ŝ12 = Ŝc22 Ŝc12 u δE
ˆ ˆ
+ Ŝc11 Ŝc12 u δE
ˆ
+ Ŝc12 u δE +
22 11 01
h  R i2 h  i2 h  i2
u δEˆ + Ŝc12 u L̂

+ Ŝc12 u Ĉ23

+
03
h  i2 h  i2
c c
Ŝ12 Ŝ22 u Ĉ22 + u R̂S12 +
  2 h  F i2 h  R i2 
c R
Ŝ Ê 1 − Ŝc Ŝc

u δE ˆ + u δE ˆ .
12 11 21 12 00 22

(H.4b)
Equations (H.4a) and (H.4b) each have one additional term, containing estimates of the
load matches, ÊF22 and ÊR
11 , of the respective ports.

H.5 Uncertainty contributions


The determination of the uncertainty terms in equations (H.2), (H.3) and (H.4) is de-
scribed below for each contribution.
H.5.1 Directivity
The uncertainties u(δE ˆ F ) and u(δE
ˆ 00 ), u(δE ˆ R ) are determined experimentally by mea-
00 33
suring the size of the residual directivity with the help of a ripple assessment. Typical
values of residual directivity are in the range of 0.001 to 0.01 (i.e., from -60 dB to -40 dB)
[67].
The procedure below is limited to one-port measurements and the determination of
ˆ 00 ). For two-port measurements the same procedure is performed at each port to
u(δE
determine u(δE ˆ F ) and u(δEˆ R ).
00 33

H.5.1.1 Ripple magnitude


Connect the air-dielectric line to the VNA test port and terminate it with a suitable low-
reflecting load. Make a measurement of Sc11 and display linear magnitude versus fre-
quency. The display should show a significant undulation, called the ’ripple’, super-
imposed on the reflection coefficient of the load itself. An example is shown in figure
H.1.
The size of the residual directivity, |δE00 |, is estimated from the peak-to-peak mag-
nitude AD of the ripple. AD should be obtained from an adjacent peak and trough with

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 94 –
0.035

0.03

0.025

0.02
|S11 |

0.015

0.01

0.005

0
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure H.1: The ripple pattern of a 30 cm beadless Type-N air-dielectric line terminated
with a matched load.

adjustment made for any slope caused by the variation with frequency of the reflection
coefficient of the terminating load, as illustrated in figure H.2. AD corresponds to an
estimate of twice the size of |δE00 |.
NOTE — The load must have a reflection coefficient ΓL that is larger than the magnitude
of the residual directivity, i.e. |ΓL | > |δE00 |. If this is not the case, the peak-to-peak ripple value
AD will be an estimate of 2 |ΓL |, rather than 2 |δE00 |. This would lead to an underestimation of
the residual directivity.

H.5.1.2 Influence of connector


As pointed out previously, the reflections ΓCO at the connector interfaces of the air-
dielectric line interfere with the ripple caused by the residual directivity. This effect has
to be considered in the uncertainty evaluation. The magnitude of the reflection can be

Figure H.2: A practical way to compensate for the slope when the peak-to-peak ripple
value AD is determined. AD is indicated by the red solid line.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 95 –
analytically approximated [68] by

|ΓCO | ' kf (1 + 0.1dpg ) (H.5)

with dpg the maximal pin gap in µm and f the frequency in GHz. The factor k is
dependent on the connector type, k = 8 · 10−5 for Type-N and k = 7 · 10−5 for 3.5 mm.
dpg is the absolute value of the sum of the recession of the center conductor at the test
port and the maximum recession of the center conductor of the air-dielectric line. The
uncertainty associated with |Γ̂CO | can be approximated by
   
u Γ̂CO ' 0.1kf · u dˆpg . (H.6)

Example: The recession of the center conductor is -20 µm at the test port, -20 µm at
the air-dielectric line and -10 µm at the matched load. This leads to a maximal pin gap of
dˆpg = 50 µm. For a measurement in Type-N at 18 GHz this leads to |Γ̂CO | ' 8.64 · 10−3
when using equation (H.5). A single measurement of the pin depth is assumed to have an
associated uncertainty of 4 µm. The addition of the three measurements leads to u(dˆpg ) =
12 µm, assuming that they are strongly correlated because the same connector gauge is used
for all three measurements. This leads to u(|Γ̂CO |) ' 1.73 · 10−3 when using equation (H.6).

NOTE — Equation (H.5) seems to favor a small maximal pin gap to keep the interfering
reflections as small as possible. However, caution is advised. Equation (H.5) takes into account
inductive coupling only. In case of a small pin gap capacitive coupling can become dominant
leading to stability problems, see 8.5.2.2. The size of this effect can’t be easily estimated an-
alytically. It is therefore necessary to repeat the ripple evaluation under different connector
orientations and record the variation in the ripple amplitude as described in H.5.1.4.

H.5.1.3 Characteristic impedance


The magnitude of the ripple is also influenced by the characteristic impedance of the
air-dielectric line. A deviation from the ideal 50 Ω leads to a reflection coefficient
Z0 − 50 Ω
|ΓAL | = (H.7)
Z0 + 50 Ω
with Z0 as defined in equation (H.1). Furthermore, there is an uncertainty associated
with the estimate of |ΓAL |, which can be approximated by
v
u   2    2
u u dˆ u dˆc
  u o
u Γ̂AL = 0.6t  + (H.8)

dˆo dˆc

with do and dc defined in H.3.

H.5.1.4 Repeatability
If more than one suitable air-dielectric line is available, the procedure should be re-
peated using the other available air-dielectric lines to check for the consistency of the
value obtained for the residual directivity. The technique should also be repeated sev-
eral times using the same air-dielectric line to estimate the variability in the value ob-
tained for the residual directivity. Typically only a small number of repetitions will be

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 96 –
done. Because of the previously mentioned effects related to capacitive coupling, it is
recommended to take a conservative approach and take the half width of the observed
variations as u(R̂E00 ).
H.5.1.5 Combined uncertainty associated with residual directivity
The uncertainty associated with an estimate of the residual directivity term can be writ-
ten as
 2  2
i2  A 2 Γ̂ Γ̂

h 
D
CO h  i 2 AL
u δEˆ 00 = √ +  √  + u Γ̂CO

+ √  +
2 2 2 2 (H.9)
h  i2 h  i2 h  i2
u Γ̂AL + u D̂00 + u R̂E00 .

The evaluations of the different terms in equation (H.9) are explained in H.5.1.1, H.5.1.2,
H.5.1.3 and H.5.1.4, respectively. The uncertainty associated √ with an estimate of the
drift u(|D̂00 |) is evaluated according to G.3. The divisor of 2 acknowledges the fact
that the phase is unknown, see [50, 69].
H.5.2 Source match
The uncertainties u(δE ˆ F ) and u(δE
ˆ 11 ), u(δE ˆ R ) are determined according to the evalua-
11 22
tion of uncertainties associated with the residual directivity, except that a high-reflecting
short is used instead of a matched load as termination of the air-dielectric line. The
residual source match is typically in the range 0.001 to 0.030 (i.e., from -60 dB to -
30 dB) [67].
The procedure below is limited to one-port measurements and the determination of
ˆ
u(δE11 ). For two-port measurements the same procedure is performed at each port to
determine u(δE ˆ F ) and u(δE ˆ R ).
11 22

H.5.2.1 Ripple magnitude


The VNA should display (again in linear magnitude format, after suitable scaling) an
increase of line attenuation with frequency, superimposed by a ripple, the amplitude
of which typically increases with frequency (though not necessarily monotonically). An
example is shown in figure H.3. An estimate of the size of the residual source match,
|δE11 |, is obtained from the peak-to-peak ripple value AS , following the same procedure
used for the directivity.
The influences due to connectors and the characteristic impedance are the same
as for the directivity and results from H.5.1.2 and H.5.1.3 can be used for the combined
uncertainty.
AS is also influenced by residual directivity, δE00 , and residual reflection tracking,
δE01 , since the ripple of this measurement includes effects from all three error coeffi-
cients. The effect of residual directivity can be taken into account using the results of
the evaluation in H.5.1.1. However, there is no suitable way to take the influence of
residual reflection tracking into account.
AS is also reduced by the losses of the air-dielectric line. The effect can be esti-
mated from the decline of the average reflection coefficient in the ripple measurement
and taken into account in the combined uncertainty by multiplying with a factor cL .

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 97 –
1

0.99

0.98

0.97

0.96
|S11 |

0.95

0.94

0.93

0.92

0.91

0.9
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure H.3: Ripple pattern of a 30 cm beadless Type-N air-dielectric line terminated with
a short.

Example: The attenuation of AS due to losses can be estimated from the average re-
flection coefficient in figure H.3. At the maximum frequency the average reflection coefficient
is approximately 0.915. Because the observed residual source match scales with S211 the ob-
served AS needs to be multiplied with a factor cL = 1/0.9152 = 1.195 at this frequency.

H.5.2.2 Combined uncertainty associated with source match


The uncertainty associated with the residual source match can be written as
 2  2
i2  A 2  A 2 Γ̂CO Γ̂

h 
S D
h  i2 AL
ˆ
u δE11 = cL √ + √ +  √  + u Γ̂CO

+  √  +
2 2 2 2 2 2
h  i2 h  i2 h  i2
u Γ̂AL + u D̂11 + u R̂E11 .

(H.10)

The first term in equation (H.10) is determined according to H.5.2.1 with AS the peak-
to-peak value in the ripple evaluation and cL the amplification factor to compensate for
losses in the air-dielectric line. The second term acknowledges the influence of the
residual directivity with AD determined in H.5.1.1. For the evaluation of the remaining
terms the same procedures as described for the residual directivity (H.5.1.2, H.5.1.3
and H.5.1.4) can be applied. The uncertainty associated √ with the estimate of drift,
u(|D̂11 |), is evaluated according to G.3. The divisor of 2 takes into account that the
phase is unknown, see [50, 69].

H.5.3 Reflection tracking


The residual reflection tracking can be characterized using reference and measurement
data of the short that has been used during VNA calibration. No additional measure-
ment standard or extra measurement is needed. However, the short must be calibrated,

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 98 –
i.e. an estimate of the reflection coefficient Γs and its associated uncertainty should be
known.
The one-port residual model, see F.5, is solved for the residual reflection tracking
coefficient δE01 while ignoring the terms related to cable, connector ,non-linearity and
noise. This leads to
(Γcs − δE00 ) (1 − δE11 Γs )
δE01 = − 1, (H.11)
Γs
with Γcs denoting the error corrected reflection coefficient of the short.
The uncertainty associated with the estimate of δE01 is calculated from uncertainties
associated with estimates of δE00 , δE11 and Γs , using linear uncertainty propagation
through equation (H.11) with best estimates δE00 = δE11 = 0 and Γ̂cs = Γ̂s . This leads
to
   2    2
h  i2 u Γ̂s u δEˆ 00 h  i2 h  i2
ˆ 01
u δE =   +   + Γ̂s u δE
ˆ 11 + u D̂01

. (H.12)
Γ̂s Γ̂s

u(δEˆ 00 ) and u(δE


ˆ 11 ) are determined by the ripple assessment described in H.5.1 and
H.5.2, respectively. The uncertainty associated with the estimate of the drift, u(|D̂01 |),
is evaluated according to G.3.
Assuming that the characterization data of the short contains uncertainties associ-
ated with the real and imaginary component estimates, perhaps even including correla-
tion, it can be simplified as
    h i  h i
u Γ̂s = max u Re Γ̂s , u Im Γ̂s . (H.13)

The evaluation according to equation (H.12) is frequency dependent, because the


characterization of the short is usually frequency dependent. For a simplified treatment
enclosing uncertainty envelopes can be defined for different frequency ranges.
Equation (H.12) leads to a large overestimation of the uncertainty associated with
the residual reflection tracking. The situation can be improved if measurement and
reference data of the calibration open are also taken into account [70]. Derivation of the
formalism is based again on equation (H.11). Combining the information about short
and open leads to
     2
h  i2 1 u Γ̂s u Γ̂o |x|2 h  ˆ i2 h  ˆ i2
 
ˆ
u δE01 = + + u δE11 + u δE00 +
 
4 Γ̂ Γ̂ 4 (H.14)
s o
h  i2
u D̂01 ,

with x denoting
Γ̂o
x=−− 1.
Γ̂s
The subscript o is referring to the calibration open. To apply equation (H.14) it is nec-
essary that short and open have well matched electrical lengths to keep the factor x

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 99 –
small. In addition it is necessary that both standards, open and short, have been char-
acterized with the same VNA in the same calibration. Both criteria are usually met if
both standards are from the same SOL calibration kit, which has been characterized as
a whole by one laboratory. If this is the case, equation (H.14) should be used instead
of (H.12).
For two-port measurements, the procedure has to be performed separately at port
one and at port two to determine u(δE ˆ F ) and u(δE
ˆ R ), respectively.
01 32

H.5.4 Transmission tracking and load match


These contributions are only relevant for two-port measurements.
In the case of SOLT calibration the measurement of the flush thru is used to relate
the one-port error coefficients with the transmission tracking and load match coeffi-
cients. This interrelation is used to estimate the uncertainties associated with resid-
ual transmission tracking and residual load match. The forward and reverse residual
two-port models are evaluated, assuming an ideal thru connection. This leads to the
following relationships for the residual transmission tracking
h  F i2 h  i2 h  i2 h  i2 h  i2
ˆ
u δE = u Ĉ32 + u L̂ + u Ŝt21 + u D̂F32 (H.15)
32
h  R i2 h  i2 h  i2 h  i2 h  i2
ˆ
u δE = u Ĉ23 + u L̂ + u Ŝt12 + u D̂R , (H.16)
01 01

and for the residual load match


h  F i2 h  F i2 h  i2 h  i2 h  i2 h  i2
u δE ˆ = u ˆ
δE + u L̂ + u Ĉ22 + u Ŝt
+ u D̂ F
22 00 11 22
(H.17)
h  R i2 h  R i2 h  i2 h  i2 h  i2 h  i2
ˆ
u δE ˆ
= u δE + u L̂ + u Ĉ33 + u Ŝt22 + u D̂R .
11 33 11
(H.18)

Uncertainties associated with estimates of δEF00 , δER R


33 , Dij , Dij , Cij and L are deter-
mined as described in H.5.1, H.5.6, H.5.7 and H.5.8.
The terms u(Ŝtij ) denote uncertainties associated with the estimates of the flush
thru measurement during calibration. The size of the uncertainty is determined by the
repeatability of the measurements. Several measurements of the flush thru have to be
performed and statistically evaluated.
H.5.5 Isolation
This contribution only applies to two-port measurements. As discussed in 5.3.5 the
effect might be of minor importance for coaxial measurements using a modern VNA.
The effect might be masked by noise floor, which is already covered in H.5.9. If this is
the case no separate uncertainty contribution needs to be specified. For all other cases
refer to 5.3.5.
H.5.6 Drift
The uncertainties associated with the estimates of drifts of the error coefficients are
determined by one of the procedures described in G.3. The uncertainty contributions

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 100 –
are added in quadrature to the corresponding uncertainties associated with the error
coefficients, as determined in H.5.1, H.5.2, H.5.3 and H.5.4.

H.5.7 Test port cables


The measurement models (H.2) and (H.3) require the test port cables to be kept fixed
on the port one side. Therefore there are only uncertainty contributions for two-port
measurements. These contributions are determined by the procedure specified in G.4.
The following relations apply.
         
ˆ ˆ
u Ĉ32 = u Ĉ23 = max u CA T , u arg CAT
     
u Ĉ22 = u Ĉ33 = u CA ˆ R .

The procedures to assign uncertainties associated with the estimates of |CAT |, arg(CAT )
and CAR are given in G.4. For arg(CAT ) the uncertainty needs to be expressed in ra-
dians.

H.5.8 Non-linearity
The uncertainty associated with the estimate of L is determined according to procedure
G.2.

H.5.9 Repeatability
This uncertainty contribution includes all random effects related to the measurement of
the DUT. It is generally dependent on the settings of the VNA, influencing noise floor
and trace noise, and the connection between test port and DUT. It is possible to quote
an uncertainty associated with the estimate of repeatability based on experience with
DUTs of similar reflection and transmission coefficient and of similar type of construc-
tion. For this purpose pooled data can be analyzed according to the formalism and the
advice given in C.5. Based on that approach, a value can be assigned to uncertainties
associated with estimates of RSij . It is nevertheless always necessary to perform at
least four measurements at different connector orientations to verify the quoted uncer-
tainty, as pointed out in 8.8.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 101 –
I Waveguide measurements
I.1 Introduction
The principles for evaluating VNA S-parameter measurements made in air-filled rect-
angular metallic waveguide are the same as for coaxial line. The measurements are
affected by the same type of measurement errors, as discussed in section 5, and the
same measurement models, as explained in appendix F, can be applied. With prop-
erly characterized calibration standards it is possible to use the Rigorous Method, see
7.1, to evaluate measurement uncertainties in the same way as it is done for coax-
ial measurements. The procedures to evaluate the basic uncertainty contributions, as
presented in appendix G, might need some adjustments to accommodate setup and
limitations specific to waveguide measurements.
This appendix gives a procedure for evaluating uncertainties following the principles
of the Ripple Method described for the coaxial case in 7.2 and appendix H. The proce-
dure is expected to be applicable to measurements made at frequencies up to 110 GHz,
or thereabouts, where mismatches due to the dimensional tolerances of the waveguide
apertures and flanges are likely to be relatively small. At frequencies above 110 GHz,
where the dimensions of the waveguide become relatively small and alignment of the
waveguide becomes more challenging, the procedure may need to be modified.

I.2 Equipment
I.2.1 VNA test ports
In order to make measurements in waveguide, the VNA needs to be configured so that
it has waveguide test ports of the correct aperture size and flange type for the required
measurements. Depending on the type of VNA, there are two ways to achieve this:

• Using coaxial cables and/or adapters to transform from coaxial tests ports found
either on the front panel of the VNA or on external coaxial Extender Heads (e.g.
as often used for 1 mm coaxial connector VNA test ports).

• Using external waveguide Extender Heads (also sometimes called Converters),


which are typically available at frequencies from 40 GHz to 110 GHz (and above).

In both cases, it is very important that the waveguide test ports consist of good quality
waveguide that is in good condition.

I.2.2 Calibration kits


At least one calibration kit is required for each waveguide size and flange type. It is
important that the standards in the calibration kit can be defined in the VNA. This is
often achieved using calibration kit definitions data (usually made available on digital
media such as disks or memory sticks) supplied by the calibration kit manufacturer.
Details of the types of calibration, and the components used for each type, are found in
I.3.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 102 –
I.2.3 Components
There are three types of components needed to evaluate the uncertainty in VNA mea-
surements using the Ripple Method in waveguide. These are described below.
At least one well-matched precision waveguide line is needed for each waveguide
size and flange type. These sections of waveguide are analogous with the beadless
reference air-dielectric lines used for the Ripple Method in coaxial line, see H.3. The
waveguide line is used for performing ‘ripple’ assessments of two post-calibration VNA
residual error terms: directivity (analogous with H.5.1 for coaxial line) and source match
(analogous with H.5.2 for coaxial line). The flange alignment and aperture dimensions
of the waveguide line should be taken into account, as part of the uncertainty assess-
ment. This is analogous with the allowance for connector interfaces and character-
istic impedance of a beadless coaxial air-dielectric line as described in H.5.1.2 and
H.5.1.3, respectively. For waveguide, the tolerances on the flange alignment mecha-
nisms should be taken into account and these should be used to calculate the worst-
case magnitude of the linear reflection coefficient (e.g. using [71, 72]). This is similar
to the evaluation of ΓCO in H.5.1.2, for coaxial line. In addition, the dimensions of the
height and width of the waveguide aperture should be measured mechanically and de-
partures from nominal dimensions should be used to calculate the magnitude of the
linear reflection coefficient, with uncertainty (e.g. using [71, 72]), analogous with ΓAL in
H.5.1.3, for coaxial line. These values of reflection coefficient should be used as part of
the uncertainty budget for the VNA’s reflection measurements in the same way as used
for coaxial measurements, analogous with equations (H.9) and (H.10).
Attenuation devices can be used to assess the non-linearity of the VNA test set
(following the procedure given in 5.3.3). In the case where the VNA test set is (or
the Extender Heads are) fitted with coaxial connectors, the non-linearity assessment
can be performed in the usual way, using coaxial attenuation devices (including step
attenuation devices). In which case, no waveguide attenuation devices are needed. For
waveguide Extender Heads (that are not assessed using coaxial attenuation devices),
it is important that the non-linearity of the Extender Heads is assessed and so it is
usual to use waveguide attenuation devices for this purpose. These attenuation devices
can either be fixed (or switched) attenuation devices, or variable attenuation devices
that can be set to fixed values, e.g. Rotary Vane Attenuation devices or Waveguide
Below Cut-Off (WBCO) attenuation devices. At millimeter-wave frequencies, another
possible means of realizing fixed values of attenuation is through the use of a cross-
connected section of a waveguide [73] (also known as highly reflective devices [74]).
These devices may be available from some waveguide manufacturers.
It is recommended that at least one verification kit is used for each waveguide size
and flange type, for the purposes of verification, see section 4. Manufacturers’ verifica-
tion kits usually contain:

• A well-matched precision waveguide line;

• A mismatched precision waveguide line. The mismatch is often produced by in-


cluding a section of reduced-height waveguide within the device;

• Two fixed attenuation devices of typically 20 dB, and, 40 dB or 50 dB.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 103 –
The well-matched precision waveguide line in the verification kit can also be used as the
precision waveguide line used for the ‘ripple’ assessments. The two fixed attenuation
devices can also be used as part of the set used for the non-linearity assessment, if the
non-linearity assessment is being performed using waveguide attenuation devices.

I.3 Calibration
I.3.1 SSL calibration
A common one-port calibration technique is the flush-short / offset-short / well-matched
load technique. This is sometimes called the SSL calibration technique (short-short-
load).
The SSL calibration technique is analogous with the SOL (short-open-load) calibra-
tion technique that is used frequently in coaxial line. Sometimes, the SSL calibration
is implemented using two offset-shorts (i.e. by using a second offset-short in place of
the flush-short). This works well as long as the phases of the reflection coefficients
produced by the offset-shorts do not coincide with each other at any frequency across
the waveguide band.
SSL calibrations can be extended to two-port calibrations by applying a SSL calibra-
tion to each of the VNA’s two test ports followed by a Thru connection (made by joining
the two waveguide tests ports together). This produces a SSLT calibration (short-short-
load-thru).

I.3.2 TRL calibration


A common two-port calibration technique in waveguide is the thru-reflect-line (TRL)
technique [48].
The TRL calibration technique is usually implemented using a length of precision wave-
guide line that is approximately a 1/4 wavelength long around the mid-band frequency
of the waveguide. This line is called the line standard. The line standard will then
produce an adequate phase change with respect to the thru standard across the full
waveguide band. The reflect standard is usually produced using a flush-short, although
other devices can be used (e.g. an offset-short).
Sometimes, at millimeter-wave frequencies (i.e. above 30 GHz), the LRL (line-reflect-
line) technique [45] is used in place of the TRL technique (i.e. by using a second line
in place of the thru). This works well if the phase difference between the two LRL line
standards is approximately a 1/4 wavelength (i.e. 90◦ ) around the mid-band frequency
of the waveguide. This technique is useful as it can avoid the need to use very short
lengths of line at high frequencies (where the wavelength is relatively short, so a 1/4
wavelength line can be very short and hence can be damaged easily).
Another method of avoiding the need to use short (1/4 wavelength) TRL line stan-
dards at millimeter-wave frequencies is to use 3/4 wavelength TRL lines. However, this
requires the use of two separate 3/4 wavelength Line standards (in conjunction with the
thru standard) in order to cover the full frequency range of the waveguide band. More
details about using 3/4 wavelength line standards for TRL calibrations is given in [75].

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 104 –
I.4 Uncertainty Evaluation
The procedure for evaluating uncertainty in waveguide VNA measurements using the
Ripple Method is very similar to the procedure in coaxial line, as explained in appendix
H. Equations to determine the uncertainties associated with estimates of S-parameters
of a DUT are as in the coaxial case (H.2) and (H.4).

I.4.1 Ripple Assessments


To evaluate the residual directivity, a ‘ripple’ assessment is performed using the preci-
sion waveguide line referred to in I.2.3, terminated with a low-reflecting load. The load
should provide a reflection coefficient with linear magnitude typically in the range 0.1 to
0.2.
An example residual directivity ‘ripple’ trace obtained for X-band waveguide (8.2
GHz to 12.4 GHz) is shown in figure I.1. This is analogous to the ripple trace shown in
figure H.1 for coaxial line.
To evaluate the residual source match, a ‘ripple’ assessment is performed using the
precision waveguide line referred to in I.2.3, terminated with a short (or offset-short).
An example residual source match ‘ripple’ trace obtained for X-band waveguide is
shown in figure I.2. This is analogous to the ripple trace shown in figure H.3 for coaxial
line.

I.4.2 Other uncertainty components


The evaluation of the other components to the uncertainty in measurements of wave-
guide devices should follow the same techniques given for measurements of coaxial
devices, as described in sub-sections of H.5.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 105 –
−3
x 10
11

10

7
|S11 |

2
8.5 9 9.5 10 10.5 11 11.5 12
Frequency /GHz

Figure I.1: Example of a residual directivity ‘ripple’ trace obtained from a VNA calibration
in X-band waveguide (operating from 8.2 GHz to 12.4 GHz).

1.005

0.995
|S11 |

0.99

0.985

0.98
8.5 9 9.5 10 10.5 11 11.5 12
Frequency /GHz

Figure I.2: Example of a residual source match ‘ripple’ trace obtained from a VNA cali-
bration in X-band waveguide (operating from 8.2 GHz to 12.4 GHz).

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 106 –
J Examples
The following examples are based on measurements of one-port and two-port DUTs in
the Type-N line system. The VNA has been calibrated with an SOLT scheme, using the
definitions provided by the manufacturer of the calibration kit. The setup was such that
test port one was kept fixed during all measurements. Therefore cable movements only
occurred during two-port measurements and did not affect the results of the one-port
DUTs. See as well 8.5.1 in that respect. The application of the Ripple Method requires
one port to be kept fixed as explained in H.4.1 and H.4.2. The following DUTs have
been measured: matched load, mismatch, short, adapter, 20 dB and 50 dB attenua-
tion device. Uncertainties have been evaluated with rigorous uncertainty propagation
through a measurement model, according to 7.1, and the Ripple Method, according to
7.2.
For the rigorous uncertainty propagation the software VNA Tools [19] has been
used. The uncertainties assigned to the calibration standards are based on manu-
facturer specifications. The other uncertainty contributions were determined according
to the procedures in appendix G. The measurement model is very similar to the 10-term
model shown in appendix F. Details of the actual measurement model being used can
be found in the documentation [19]. The uncertainty contributions are propagated to
the DUT results using linear uncertainty propagation. The results below show measure-
ment uncertainty intervals for single components of the complex quantities. However,
the method also evaluates the correlation between S-parameters as well as the inter
frequency correlations, but this is not shown.
For the Ripple Method a 30 cm long beadless air-dielectric line was used together
with a matched load and a short. The center conductors showed a recession of -
67.8 µm (still within IEEE specifications [65]) for the test port, -4.8 µm for the air-dielectric
line and -2.5 µm for the matched load. The short had a slight protrusion of +2.8 µm. The
separate sums of these values for the configurations with the matched load and with the
short are needed to evaluate uncertainty contributions of directivity and source match
according to equation (H.5). That calculation leads to dpg = 75.1 µm for the matched
load and dpg = 69.8 µm for the short. The uncertainty associated with a single mea-
surement of the pin depth is 3.8 µm. Adding three measurements and assuming full
correlation leads to u(dpg ) = 11.4 µm, which is needed in equation (H.6). Figure J.1 il-
lustrates four measurements of the ripple pattern taken with the matched load. Only the
longitudinal position of the center conductor of the air-dielectric line was varied, using
the whole available space. The two extreme positions were achieved by pushing the
inner conductor onto the test port and onto the load before making the connection. The
two other positions were achieved with the help of dielectric spacers. This illustrates the
effect of the connector interfaces on the ripple pattern. The blue trace in figure J.1 has
been used later in the examples to determine the uncertainty associated with residual
directivity.
The uncertainties associated with the DUT are evaluated according to the procedure
described in appendix H. The uncertainties quoted in the examples are all standard un-
certainties (k = 1). The examples show results for S11 for the one-port devices and
S22 and S12 for the two-port devices. The results are representative of S11 and S21 as

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 107 –
0.04

0.035

0.03

0.025
|S11 |

0.02

0.015

0.01

0.005

0
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.1: Four different ripple patterns in the Type-N line system. All four traces were
taken with a 30 cm beadless air-dielectric line terminated with a matched load. The dif-
ferences are due to different longitudinal positions of the center conductor of the air-
dielectric line. See also H.5.1.2

well. Results are expressed either in real and imaginary components, for low reflection
coefficients, or magnitude and phase, for the other cases. The magnitude of the trans-
mission coefficient is displayed in logarithmic units. The results are shown graphically
over the whole frequency range. For three selected frequencies, uncertainty budgets
are shown in tabular form. The uncertainty budget of the Ripple Method quotes a sin-
gle uncertainty u(Sij ) = u(Re[Sij ]) = u(Im[Sij ]), as explained in H.2. The uncertainties
associated with magnitude and phase are derived using linear uncertainty propagation

u(|Sij |) = u(Sij )
u(Sij )
u(arg (Sij )) = .
|Sij |

The transformation between linear and logarithmic units is also done using linear un-
certainty propagation

20 u(Xlin ) u(Xlin )
Xlog = 20 log10 Xlin −→ u(Xlog ) = ' 8.686
loge 10 Xlin Xlin
Xlog u(Xlin ) loge 10
Xlin = 10 20 −→ = u(Xlog ) ' 0.1151u(Xlog ) .
Xlin 20
The numerical values of the uncertainty contributions in these examples should not
be taken as generally representative. E.g. the uncertainty associated with the repeata-
bility of the DUT measurements is very small in these examples. Such uncertainties
can only be achieved with appropriate device settings, a well controlled measurement
process and stable environmental conditions. It is necessary to individually character-
ize the uncertainty contributions for the specific measurement setups. This can also be

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 108 –
combined with experience and other sources of information, as e.g. manufacturer spec-
ifications, if available. For contributions that are not dominant, a conservatively large
uncertainty value might be assigned without affecting the final combined uncertainty
too much. This way it is possible to reduce the characterization effort.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 109 –
J.1 One-port matched load

0.015

0.01

0.005

0
Re[S11 ]

−0.005

−0.01

−0.015

−0.02
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.2: Real component of the reflection coefficient of a matched load with associ-
ated standard uncertainties. The solid lines indicate the error corrected measurement
data (center line) bounded by the interval of one standard uncertainty (k = 1) calculated
with the rigorous method. The step in uncertainty at 2 GHz is due to the change from
low band load to sliding load. The vertical bars indicate standard uncertainties evaluated
with the Ripple Method.

0.02

0.015

0.01

0.005
Im[S11 ]

−0.005

−0.01

−0.015

−0.02
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.3: The same as figure J.2 but for the imaginary component of the reflection
coefficient.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 110 –
Table J.1: Uncertainty budget of Re [S11 ] of a matched load at three different frequencies.
The standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00200 0.00398 0.00398


Cal Open 0.00000 0.00004 0.00005
Cal Short 0.00000 0.00004 0.00005
Conn. Rep. 0.00045 0.00071 0.00071
VNA Drift 0.00002 0.00004 0.00004
VNA Linearity 0.00001 0.00003 0.00009
VNA Noise 0.00002 0.00001 0.00002

Combined 0.00205 0.00404 0.00404

Table J.2: The same as table J.1 but for Im [S11 ]

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00200 0.00398 0.00398


Cal Open 0.00000 0.00004 0.00005
Cal Short 0.00000 0.00004 0.00005
Conn. Rep. 0.00045 0.00071 0.00071
VNA Drift 0.00002 0.00004 0.00004
VNA Linearity 0.00001 0.00003 0.00008
VNA Noise 0.00002 0.00001 0.00002

Combined 0.00205 0.00404 0.00404

Table J.3: Uncertainty budget of S11 of a matched load at three different frequencies. The
standard uncertainty is evaluated with the Ripple Method. The terms in column 2 refer
to equation (H.2) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Directivity u(δE ˆ 00 ) 0.00228 0.00540 0.01103


Reflection tracking ˆ 01 )
|Ŝc11 |u(δE 0.00001 0.00007 0.00009
Source match c 2
|Ŝ | u(δE ˆ 11 ) 0.00001 0.00001 0.00001
11
Linearity |Ŝc11 |u(L̂) 0.00001 0.00001 0.00001
Repeatability u(R̂S11 ) 0.00100 0.00100 0.00100

Combined uncertainty u(Ŝ11 ) 0.00249 0.00549 0.01107

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 111 –
J.2 One-port mismatch

0.23

0.22

0.21

0.2
|S11 |

0.19

0.18

0.17

0.16
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.4: Magnitude of the reflection coefficient of a mismatch with associated stan-
dard uncertainties. The solid lines indicate the error corrected measurement data (center
line) bounded by the interval of one standard uncertainty (k = 1) calculated with the rigor-
ous method. The step in uncertainty at 2 GHz is due to the change from low band load to
sliding load. The vertical bars indicate standard uncertainties evaluated with the Ripple
Method.

1
arg (S11 ) /◦

−1

−2

−3

−4
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.5: The same as figure J.4 but for the phase of the reflection coefficient. For
better visibility the error corrected measurement has been normalized to 0 and only the
standard uncertainties are shown.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 112 –
Table J.4: Uncertainty budget of |S11 | of a mismatch at three different frequencies. The
standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00194 0.00393 0.00411


Cal Open 0.00041 0.00119 0.00310
Cal Short 0.00060 0.00162 0.00244
Conn. Rep. 0.00045 0.00071 0.00073
VNA Drift 0.00003 0.00003 0.00003
VNA Linearity 0.00016 0.00013 0.00017
VNA Noise 0.00003 0.00002 0.00003

Combined 0.00212 0.00446 0.00574

Table J.5: The same as table J.4 but for arg (S11 )

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.5493 deg 1.1839 deg 1.1198 deg


Cal Open 0.1148 deg 0.3568 deg 0.8434 deg
Cal Short 0.1679 deg 0.4868 deg 0.6656 deg
Conn. Rep. 0.1265 deg 0.2142 deg 0.1985 deg
VNA Drift 0.0081 deg 0.0064 deg 0.0064 deg
VNA Linearity 0.0394 deg 0.0332 deg 0.0396 deg
VNA Noise 0.0054 deg 0.0032 deg 0.0075 deg

Combined 0.6006 deg 1.3464 deg 1.5650 deg

Table J.6: Uncertainty budget of S11 of a mismatch at three different frequencies. The
standard uncertainty is evaluated with the Ripple method. The terms in column 2 refer
to equation (H.2) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Directivity u(δE ˆ 00 ) 0.00228 0.00540 0.01103


Reflection tracking ˆ 01 )
|Ŝc11 |u(δE 0.00103 0.00279 0.00541
Source match c 2
|Ŝ | u(δE ˆ 11 ) 0.00042 0.00043 0.00081
11
Linearity |Ŝc11 |u(L̂) 0.00027 0.00025 0.00028
Repeatability u(R̂S11 ) 0.00100 0.00100 0.00100

Combined uncertainty u(Ŝ11 ) 0.00274 0.00618 0.01235

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 113 –
J.3 One-port short

1.04

1.03

1.02

1.01
|S11 |

0.99

0.98

0.97

0.96
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.6: Magnitude of the reflection coefficient of a short with associated standard
uncertainties. The solid lines indicate the error corrected measurement data (center line)
bounded by the interval of one standard uncertainty (k = 1) calculated with the rigor-
ous method. The vertical bars indicate standard uncertainties evaluated with the Ripple
Method.

1.5

0.5
arg (S11 ) /◦

−0.5

−1

−1.5

−2
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.7: The same as figure J.6 but for the phase of the reflection coefficient. For
better visibility the error corrected measurement has been normalized to 0 and only the
standard uncertainties are shown.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 114 –
Table J.7: Uncertainty budget of |S11 | of a short at three different frequencies. The stan-
dard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00001 0.00002 0.00004


Cal Open 0.00001 0.00002 0.00006
Cal Short 0.00493 0.01460 0.02558
Conn. Rep. 0.00064 0.00100 0.00100
VNA Drift 0.00008 0.00008 0.00008
VNA Linearity 0.00036 0.00000 0.00000
VNA Noise 0.00012 0.00011 0.00014

Combined 0.00498 0.01463 0.02560

Table J.8: The same as table J.7 but for arg (S11 )

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.0003 deg 0.0006 deg 0.0019 deg


Cal Open 0.0002 deg 0.0006 deg 0.0031 deg
Cal Short 0.2821 deg 0.8392 deg 1.4731 deg
Conn. Rep. 0.0362 deg 0.0573 deg 0.0573 deg
VNA Drift 0.0042 deg 0.0041 deg 0.0040 deg
VNA Linearity 0.0187 deg 0.0000 deg 0.0000 deg
VNA Noise 0.0020 deg 0.0036 deg 0.0094 deg

Combined 0.2851 deg 0.8411 deg 1.4743 deg

Table J.9: Uncertainty budget of S11 of a short at three different frequencies. The stan-
dard uncertainty is evaluated with the Ripple method. The terms in column 2 refer to
equation (H.2) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Directivity u(δE ˆ 00 ) 0.00228 0.00540 0.01103


Reflection tracking ˆ 01 )
|Ŝc11 |u(δE 0.00509 0.01465 0.02562
Source match c 2
|Ŝ | u(δE ˆ 11 ) 0.01026 0.01172 0.01807
11
Linearity |Ŝc11 |u(L̂) 0.00130 0.00130 0.00130
Repeatability u(R̂S11 ) 0.00100 0.00100 0.00100

Combined uncertainty u(Ŝ11 ) 0.01179 0.01959 0.03327

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 115 –
J.4 Two-port Adapter

0.025

0.02

0.015

0.01
Re[S22 ]

0.005

−0.005

−0.01

−0.015

−0.02
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.8: Real component of the reflection coefficient of an adapter with associated
standard uncertainties. The solid lines indicate the error corrected measurement data
(center line) bounded by the interval of one standard uncertainty (k = 1) calculated with
the rigorous method. The step in uncertainty at 2 GHz is due to the change from low band
load to sliding load. The vertical bars indicate standard uncertainties evaluated with the
Ripple Method.

0.03

0.02

0.01
Im[S22 ]

−0.01

−0.02

−0.03
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.9: The same as figure J.8 but for the imaginary component of the reflection
coefficient.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 116 –
Table J.10: Uncertainty budget of Re [S22 ] of an adapter at three different frequencies.
The standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00326 0.00732 0.00684


Cal Open 0.00001 0.00006 0.00014
Cal Short 0.00001 0.00007 0.00014
Conn. Rep. 0.00082 0.00135 0.00129
Cable 0.00216 0.00231 0.00221
VNA Drift 0.00015 0.00016 0.00015
VNA Linearity 0.00002 0.00004 0.00013
VNA Noise 0.00003 0.00001 0.00003

Combined 0.00399 0.00779 0.00731

Table J.11: The same as table J.10 but for Im [S22 ]

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00326 0.00732 0.00684


Cal Open 0.00001 0.00006 0.00014
Cal Short 0.00001 0.00007 0.00014
Conn. Rep. 0.00082 0.00135 0.00129
Cable 0.00216 0.00231 0.00221
VNA Drift 0.00015 0.00016 0.00015
VNA Linearity 0.00002 0.00005 0.00013
VNA Noise 0.00003 0.00001 0.00003

Combined 0.00399 0.00779 0.00731

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 117 –
Table J.12: Uncertainty budget of S22 of an adapter at three different frequencies. The
standard uncertainty is evaluated with the Ripple Method. The terms in column 2 refer
to equation (H.3) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Directivity u(δE ˆ R) 0.00228 0.00540 0.01103


33

Reflection tracking |Ŝc22 |u(δEˆ R) 0.00002 0.00013 0.00027


32
R
Source match c 2 ˆ
|Ŝ22 | u(δE22 ) 0.00001 0.00001 0.00001
Load match |Ŝc21 Ŝc12 |u(δEˆ R) 0.00331 0.00572 0.01067
11
c
Linearity |Ŝ22 |u(L̂) 0.00001 0.00002 0.00002
Cable transmission |Ŝc22 |u(Ĉ32 ) 0.00001 0.00005 0.00009
Cable transmission |Ŝc22 |u(Ĉ23 ) 0.00001 0.00005 0.00009
Cable reflection |Ŝc22 |2 u(Ĉ22 ) 0.00001 0.00001 0.00001
Cable reflection u(Ĉ33 ) 0.00200 0.00200 0.00200
Repeatability u(R̂S22 ) 0.00100 0.00100 0.00100

Combined uncertainty u(Ŝ22 ) 0.00460 0.00817 0.01551

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 118 –
0.05

−0.05

|S12 | /dB −0.1

−0.15

−0.2

−0.25

−0.3

−0.35

−0.4
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.10: Magnitude of the transmission coefficient of an adapter with associated


standard uncertainties. The solid lines indicate the error corrected measurement data
(center line) bounded by the interval of one standard uncertainty (k = 1) calculated with
the rigorous method. The vertical bars indicate standard uncertainties evaluated with the
Ripple Method.

0.8

0.6

0.4

0.2
arg (S12 ) /◦

−0.2

−0.4

−0.6

−0.8
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.11: The same as figure J.10 but for the phase of the transmission coefficient.
For better visibility the error corrected measurement has been normalized to 0 and only
the standard uncertainties are shown.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 119 –
Table J.13: Uncertainty budget of |S12 | of an adapter at three different frequencies. The
standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.0007 dB 0.0026 dB 0.0028 dB


Cal Open 0.0006 dB 0.0035 dB 0.0072 dB
Cal Short 0.0006 dB 0.0035 dB 0.0071 dB
Conn. Rep. 0.0002 dB 0.0005 dB 0.0006 dB
Cable 0.0145 dB 0.0208 dB 0.0280 dB
VNA Drift 0.0010 dB 0.0010 dB 0.0010 dB
VNA Linearity 0.0032 dB 0.0071 dB 0.0071 dB
VNA Noise 0.0010 dB 0.0010 dB 0.0010 dB

Combined 0.0150 dB 0.0227 dB 0.0307 dB

Table J.14: The same as table J.13 but for arg (S12 )

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.0046 deg 0.0174 deg 0.0184 deg


Cal Open 0.0041 deg 0.0231 deg 0.0473 deg
Cal Short 0.0041 deg 0.0232 deg 0.0468 deg
Conn. Rep. 0.0012 deg 0.0034 deg 0.0037 deg
Cable 0.1559 deg 0.4079 deg 0.6944 deg
VNA Drift 0.0061 deg 0.0061 deg 0.0061 deg
VNA Linearity 0.0190 deg 0.0425 deg 0.0424 deg
VNA Noise 0.0020 deg 0.0037 deg 0.0081 deg

Combined 0.1573 deg 0.4118 deg 0.6992 deg

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 120 –
Table J.15: Uncertainty budget of S12 of an adapter at three different frequencies. The
standard uncertainty is evaluated with the Ripple Method. The terms in column 2 refer
to equation (H.4) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Source match |Ŝc22 Ŝc12 |u(δEˆ R) 0.0003 dB 0.0009 dB 0.0017 dB


22

Load match |Ŝc11 Ŝc12 |u(δEˆ R) 0.0001 dB 0.0005 dB 0.0010 dB


11

Transmission tracking |Ŝc12 |u(δE ˆ R) 0.0232 dB 0.0466 dB 0.0762 dB


01

Isolation u(δE ˆ R) 0.0001 dB 0.0001 dB 0.0001 dB


03
Linearity |Ŝc12 |u(L̂) 0.0113 dB 0.0113 dB 0.0113 dB
Cable transmission |Ŝc12 |u(Ĉ23 ) 0.0168 dB 0.0438 dB 0.0745 dB
Cable reflection |Ŝc12 Ŝc22 |u(Ĉ22 ) 0.0001 dB 0.0002 dB 0.0002 dB
Repeatability u(R̂S12 ) 0.0087 dB 0.0087 dB 0.0087 dB
Raw load match |Ŝc12 ÊR − ˆF)
Ŝc21 Ŝc12 )|u(δE 0.0006 dB 0.0024 dB 0.0054 dB
11 (1 00

Raw load match c


|Ŝ12 ÊR − ˆ R)
Ŝc21 Ŝc12 )|u(δE 0.0025 dB 0.0052 dB 0.0088 dB
11 (1 22

Combined uncertainty u(Ŝ12 ) 0.0364 dB 0.0678 dB 0.1089 dB

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 121 –
J.5 Two-port 20 dB attenuation device

0.06

0.04

0.02

0
Re[S22 ]

−0.02

−0.04

−0.06

−0.08

−0.1
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.12: Real component of the reflection coefficient of a 20 dB attenuation device


with associated standard uncertainties. The solid lines indicate the error corrected mea-
surement data (center line) bounded by the interval of one standard uncertainty (k = 1)
calculated with the rigorous method. The step in uncertainty at 2 GHz is due to the
change from low band load to sliding load. The vertical bars indicate standard uncertain-
ties evaluated with the Ripple Method.

0.06

0.04

0.02

0
Im[S22 ]

−0.02

−0.04

−0.06

−0.08
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.13: The same as figure J.12 but for the imaginary component of the reflection
coefficient.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 122 –
Table J.16: Uncertainty budget of Re [S22 ] of a 20 dB attenuation device at three different
frequencies. The standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00199 0.00400 0.00394


Cal Open 0.00003 0.00027 0.00114
Cal Short 0.00003 0.00029 0.00108
Conn. Rep. 0.00045 0.00071 0.00071
Cable 0.00141 0.00150 0.00172
VNA Drift 0.00009 0.00009 0.00010
VNA Linearity 0.00002 0.00002 0.00009
VNA Noise 0.00002 0.00001 0.00002

Combined 0.00248 0.00435 0.00463

Table J.17: The same as table J.16 but for Im [S22 ]

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00199 0.00400 0.00394


Cal Open 0.00003 0.00027 0.00114
Cal Short 0.00003 0.00029 0.00108
Conn. Rep. 0.00045 0.00071 0.00071
Cable 0.00141 0.00145 0.00238
VNA Drift 0.00009 0.00009 0.00010
VNA Linearity 0.00002 0.00002 0.00008
VNA Noise 0.00002 0.00001 0.00002

Combined 0.00248 0.00433 0.00491

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 123 –
Table J.18: Uncertainty budget of S22 of a 20 dB attenuation device at three different fre-
quencies. The standard uncertainty is evaluated with the Ripple Method. The terms
in column 2 refer to equation (H.3) and the corresponding values in the subsequent
columns have to be added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Directivity u(δE ˆ R) 0.00228 0.00540 0.01103


33

Reflection tracking |Ŝc22 |u(δEˆ R) 0.00005 0.00055 0.00222


32
R
Source match c 2 ˆ
|Ŝ22 | u(δE22 ) 0.00001 0.00002 0.00014
Load match |Ŝc21 Ŝc12 |u(δEˆ R) 0.00004 0.00006 0.00012
11
c
Linearity |Ŝ22 |u(L̂) 0.00002 0.00005 0.00012
Cable transmission |Ŝc22 |u(Ĉ32 ) 0.00002 0.00019 0.00074
Cable transmission |Ŝc22 |u(Ĉ23 ) 0.00002 0.00019 0.00074
Cable reflection |Ŝc22 |2 u(Ĉ22 ) 0.00001 0.00001 0.00002
Cable reflection u(Ĉ33 ) 0.00200 0.00200 0.00200
Repeatability u(R̂S22 ) 0.00100 0.00100 0.00100

Combined uncertainty u(Ŝ22 ) 0.00320 0.00587 0.01152

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 124 –
−19.7

−19.75

−19.8

−19.85
|S12 | /dB

−19.9

−19.95

−20

−20.05

−20.1
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.14: Magnitude of the transmission coefficient of a 20 dB attenuation device


with associated standard uncertainties. The solid lines indicate the error corrected mea-
surement data (center line) bounded by the interval of one standard uncertainty (k = 1)
calculated with the rigorous method. The vertical bars indicate standard uncertainties
evaluated with the Ripple Method.

0.8

0.6

0.4

0.2
arg (S12 ) /◦

−0.2

−0.4

−0.6

−0.8
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.15: The same as figure J.14 but for the phase of the transmission coefficient.
For better visibility the error corrected measurement has been normalized to 0 and only
the standard uncertainties are shown.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 125 –
Table J.19: Uncertainty budget of |S12 | of a 20 dB attenuation device at three different
frequencies. The standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.0006 dB 0.0010 dB 0.0044 dB


Cal Open 0.0006 dB 0.0041 dB 0.0063 dB
Cal Short 0.0006 dB 0.0039 dB 0.0058 dB
Conn. Rep. 0.0001 dB 0.0006 dB 0.0010 dB
Cable 0.0145 dB 0.0209 dB 0.0279 dB
VNA Drift 0.0018 dB 0.0018 dB 0.0018 dB
VNA Linearity 0.0071 dB 0.0071 dB 0.0071 dB
VNA Noise 0.0015 dB 0.0010 dB 0.0012 dB

Combined 0.0163 dB 0.0228 dB 0.0305 dB

Table J.20: The same as table J.19 but for arg (S12 )

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.0038 deg 0.0063 deg 0.0290 deg


Cal Open 0.0037 deg 0.0267 deg 0.0418 deg
Cal Short 0.0037 deg 0.0254 deg 0.0383 deg
Conn. Rep. 0.0009 deg 0.0038 deg 0.0067 deg
Cable 0.1559 deg 0.4085 deg 0.6929 deg
VNA Drift 0.0108 deg 0.0108 deg 0.0108 deg
VNA Linearity 0.0424 deg 0.0424 deg 0.0425 deg
VNA Noise 0.0077 deg 0.0045 deg 0.0091 deg

Combined 0.1622 deg 0.4126 deg 0.6973 deg

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 126 –
Table J.21: Uncertainty budget of S12 of a 20 dB attenuation device at three different fre-
quencies. The standard uncertainty is evaluated with the Ripple Method. The terms
in column 2 refer to equation (H.4) and the corresponding values in the subsequent
columns have to be added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Source match |Ŝc22 Ŝc12 |u(δEˆ R) 0.0009 dB 0.0039 dB 0.0137 dB


22

Load match |Ŝc11 Ŝc12 |u(δEˆ R) 0.0001 dB 0.0031 dB 0.0092 dB


11

Transmission tracking |Ŝc12 |u(δE ˆ R) 0.0232 dB 0.0466 dB 0.0762 dB


01

Isolation u(δE ˆ R) 0.0001 dB 0.0001 dB 0.0001 dB


03
Linearity |Ŝc12 |u(L̂) 0.0113 dB 0.0113 dB 0.0113 dB
Cable transmission |Ŝc12 |u(Ĉ23 ) 0.0168 dB 0.0438 dB 0.0745 dB
Cable reflection |Ŝc12 Ŝc22 |u(Ĉ22 ) 0.0002 dB 0.0007 dB 0.0015 dB
Repeatability u(R̂S12 ) 0.0087 dB 0.0087 dB 0.0087 dB
Raw load match |Ŝc12 ÊR − ˆF)
Ŝc21 Ŝc12 )|u(δE 0.0004 dB 0.0013 dB 0.0031 dB
11 (1 00

Raw load match c


|Ŝ12 ÊR − ˆ R)
Ŝc21 Ŝc12 )|u(δE 0.0015 dB 0.0029 dB 0.0051 dB
11 (1 22

Combined uncertainty u(Ŝ12 ) 0.0364 dB 0.0679 dB 0.1101 dB

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 127 –
J.6 Two-port 50 dB attenuation device

0.08

0.06

0.04

0.02
Re[S22 ]

−0.02

−0.04

−0.06

−0.08
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.16: Real component of the reflection coefficient of a 50 dB attenuation device


with associated standard uncertainties. The solid lines indicate the error corrected mea-
surement data (center line) bounded by the interval of one standard uncertainty (k = 1)
calculated with the rigorous method. The step in uncertainty at 2 GHz is due to the
change from low band load to sliding load. The vertical bars indicate standard uncertain-
ties evaluated with the Ripple Method.

0.06

0.04

0.02

0
Im[S22 ]

−0.02

−0.04

−0.06

−0.08

−0.1
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.17: The same as figure J.16 but for the imaginary component of the reflection
coefficient.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 128 –
Table J.22: Uncertainty budget of Re [S22 ] of a 50 dB attenuation device at three different
frequencies. The standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00199 0.00398 0.00398


Cal Open 0.00005 0.00007 0.00035
Cal Short 0.00005 0.00007 0.00036
Conn. Rep. 0.00045 0.00071 0.00071
Cable 0.00142 0.00142 0.00151
VNA Drift 0.00010 0.00009 0.00009
VNA Linearity 0.00002 0.00003 0.00011
VNA Noise 0.00002 0.00001 0.00002

Combined 0.00249 0.00428 0.00435

Table J.23: The same as table J.22 but for Im [S22 ]

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.00199 0.00398 0.00398


Cal Open 0.00005 0.00007 0.00035
Cal Short 0.00005 0.00007 0.00036
Conn. Rep. 0.00045 0.00071 0.00071
Cable 0.00142 0.00142 0.00148
VNA Drift 0.00010 0.00009 0.00009
VNA Linearity 0.00002 0.00003 0.00010
VNA Noise 0.00002 0.00001 0.00002

Combined 0.00249 0.00428 0.00434

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 129 –
Table J.24: Uncertainty budget of S22 of a 50 dB attenuation device at three different fre-
quencies. The standard uncertainty is evaluated with the Ripple Method. The terms
in column 2 refer to equation (H.3) and the corresponding values in the subsequent
columns have to be added accordingly for the combined standard uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Directivity u(δE ˆ R) 0.00228 0.00540 0.01103


33

Reflection tracking |Ŝc22 |u(δEˆ R) 0.00010 0.00014 0.00071


32
R
Source match c 2 ˆ
|Ŝ22 | u(δE22 ) 0.00001 0.00001 0.00002
Load match |Ŝc21 Ŝc12 |u(δEˆ R) 0.00001 0.00001 0.00001
11
c
Linearity |Ŝ22 |u(L̂) 0.00003 0.00002 0.00004
Cable transmission |Ŝc22 |u(Ĉ32 ) 0.00004 0.00005 0.00024
Cable transmission |Ŝc22 |u(Ĉ23 ) 0.00004 0.00005 0.00024
Cable reflection |Ŝc22 |2 u(Ĉ22 ) 0.00001 0.00001 0.00001
Cable reflection u(Ĉ33 ) 0.00200 0.00200 0.00200
Repeatability u(R̂S22 ) 0.00100 0.00100 0.00100

Combined uncertainty u(Ŝ22 ) 0.00320 0.00584 0.01128

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 130 –
−49.2

−49.4

−49.6

|S12 | /dB −49.8

−50

−50.2

−50.4

−50.6

−50.8

−51
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.18: Magnitude of the transmission coefficient of a 50 dB attenuation device


with associated standard uncertainties. The solid lines indicate the error corrected mea-
surement data (center line) bounded by the interval of one standard uncertainty (k = 1)
calculated with the rigorous method. The vertical bars indicate standard uncertainties
evaluated with the Ripple Method.

0.8

0.6

0.4

0.2
arg (S12 ) /◦

−0.2

−0.4

−0.6

−0.8
0 2 4 6 8 10 12 14 16 18
Frequency /GHz

Figure J.19: The same as figure J.18 but for the phase of the transmission coefficient.
For better visibility the error corrected measurement has been normalized to 0 and only
the standard uncertainties are shown.

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 131 –
Table J.25: Uncertainty budget of |S12 | of a 50 dB attenuation device at three different
frequencies. The standard uncertainty is evaluated with the rigorous method.

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.0005 dB 0.0014 dB 0.0021 dB


Cal Open 0.0002 dB 0.0012 dB 0.0065 dB
Cal Short 0.0002 dB 0.0012 dB 0.0066 dB
Conn. Rep. 0.0001 dB 0.0003 dB 0.0005 dB
Cable 0.0145 dB 0.0208 dB 0.0280 dB
VNA Drift 0.0028 dB 0.0028 dB 0.0028 dB
VNA Linearity 0.0071 dB 0.0071 dB 0.0071 dB
VNA Noise 0.0383 dB 0.0133 dB 0.0178 dB

Combined 0.0417 dB 0.0259 dB 0.0353 dB

Table J.26: The same as table J.25 but for arg (S12 )

Contribution 0.460 GHz 8.480 GHz 17.600 GHz

Cal Load 0.0032 deg 0.0094 deg 0.0138 deg


Cal Open 0.0011 deg 0.0076 deg 0.0426 deg
Cal Short 0.0010 deg 0.0079 deg 0.0434 deg
Conn. Rep. 0.0009 deg 0.0018 deg 0.0031 deg
Cable 0.1559 deg 0.4078 deg 0.6940 deg
VNA Drift 0.0166 deg 0.0166 deg 0.0166 deg
VNA Linearity 0.0424 deg 0.0424 deg 0.0425 deg
VNA Noise 0.2525 deg 0.0872 deg 0.1176 deg

Combined 0.3003 deg 0.4198 deg 0.7081 deg

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 132 –
Table J.27: Uncertainty budget of S12 of a 50 dB attenuation device at three different fre-
quencies. The standard uncertainty is evaluated with the Ripple Method. The terms
in column 2 refer to equation (H.4) and the corresponding values in the subsequent
columns have to be added accordingly for the combined uncertainty.

Name Uncertainty contribution 0.460 GHz 8.480 GHz 17.600 GHz

Source match |Ŝc22 Ŝc12 |u(δEˆ R) 0.0017 dB 0.0010 dB 0.0044 dB


22

Load match |Ŝc11 Ŝc12 |u(δEˆ R) 0.0005 dB 0.0007 dB 0.0026 dB


11

Transmission tracking |Ŝc12 |u(δE ˆ R) 0.0232 dB 0.0466 dB 0.0762 dB


01

Isolation u(δE ˆ R) 0.0030 dB 0.0030 dB 0.0026 dB


03
Linearity |Ŝc12 |u(L̂) 0.0113 dB 0.0113 dB 0.0113 dB
Cable transmission |Ŝc12 |u(Ĉ23 ) 0.0168 dB 0.0438 dB 0.0745 dB
Cable reflection |Ŝc12 Ŝc22 |u(Ĉ22 ) 0.0004 dB 0.0002 dB 0.0005 dB
Repeatability u(R̂S12 ) 0.0087 dB 0.0087 dB 0.0087 dB
Raw load match |Ŝc12 ÊR − ˆF)
Ŝc21 Ŝc12 )|u(δE 0.0004 dB 0.0013 dB 0.0031 dB
11 (1 00

Raw load match c


|Ŝ12 ÊR − ˆ R)
Ŝc21 Ŝc12 )|u(δE 0.0015 dB 0.0028 dB 0.0052 dB
11 (1 22

Combined uncertainty u(Ŝ12 ) 0.0365 dB 0.0678 dB 0.1091 dB

EURAMET Calibration Guide No. 12


Version 3.0 (03/2018) – 133 –
EURAMET e.V. Phone: +49 531 592 1960
Bundesallee 100 Fax: +49 531 592 1969
38116 Braunschweig E-mail: secretariat@euramet.org
Germany

EURAMET e.V. is a non-profit association under German law.

Вам также может понравиться